CRAN Package Check Results for Package PRISMA

Last updated on 2014-09-12 01:48:12.

Check Details

Version: 0.2-2
Check: R code for possible problems
Result: NOTE
    corpusToPrisma: no visible global function definition for
     ‘TermDocumentMatrix’
Flavors: r-devel-linux-x86_64-debian-clang, r-devel-linux-x86_64-debian-gcc, r-devel-linux-x86_64-fedora-clang, r-devel-linux-x86_64-fedora-gcc, r-devel-osx-x86_64-clang, r-devel-windows-ix86+x86_64

Version: 0.2-2
Check: examples
Result: ERROR
    Running examples in 'PRISMA-Ex.R' failed
    The error most likely occurred in:
    
    > ### Name: corpusToPrisma
    > ### Title: Convert tm copus to PRISMA
    > ### Aliases: corpusToPrisma
    >
    > ### ** Examples
    >
    > data(thesis)
    > thesis
    $content
    $content[[1]]
    $content
     [1] " prisma method"
     [2] ""
     [3] ""
     [4] ""
     [5] ""
     [6] " protocol inspect state machin"
     [7] " analysi prisma"
     [8] ""
     [9] ""
     [10] ""
     [11] "given collect record traffic specif network servic"
     [12] " goal prisma extract state machin associ"
     [13] "templat rule describ inform flow messag"
     [14] " messag first preprocess stage raw network"
     [15] "traffic convert session contain messag method"
     [16] "proceed follow step see figur"
     [17] ""
     [18] ""
     [19] " find common structur first defin"
     [20] " similar measur messag done"
     [21] " messag special vector space "
     [22] " reduc via statist test focus discrimin featur"
     [23] " see"
     [24] ""
     [25] " proceed model session messag sequenc"
     [26] " leverag embed previous step"
     [27] " appli cluster"
     [28] " group individu messag event see"
     [29] ""
     [30] " extract sequenc event can seen path"
     [31] " protocol state machin infer approxim"
     [32] " state machin use probabilist concept "
     [33] " model transit link probabl see"
     [34] ""
     [35] ""
     [36] " final automat generat "
     [37] " messag associ state markov deriv"
     [38] " describ inform flow "
     [39] " differ state communic see"
     [40] ""
     [41] ""
     [42] ""
     [43] "throughout chapter use term atom"
     [44] "exchang byte sequenc client server"
     [45] " describ certain action client server side"
     [46] " direct connect state machin model"
     [47] "network servic messag structur consist"
     [48] " sequenc token field describ messag"
     [49] "flow field consecut templat instanti "
     [50] "concret session"
     [51] ""
     [52] " network"
     [53] ""
     [54] ""
     [55] " learn inner structur behavior specif network"
     [56] "servic first collect suffici "
     [57] "infer normal can done dedic sensor"
     [58] "collect raw packet binari format instanc via"
     [59] "tool apart payload packet contain"
     [60] "sourc destin address actual reconstruct"
     [61] "inform flow client server packet "
     [62] " reassembl elimin artifact network transport"
     [63] "layer task devis network record use"
     [64] "matur librari reassembl tcp udp"
     [65] "communic stream"
     [66] ""
     [67] " stream input session extractor generat"
     [68] " reassembl packet payload specif session identifi"
     [69] "accord sourc destin packet two packet"
     [70] "occur small delay millisecond"
     [71] "payload will merg specif session identifi "
     [72] "sourc destin communic within"
     [73] " millisecond correspond session will"
     [74] "flag termin messag arriv "
     [75] "specif sourcedestin combin will open new session"
     [76] ""
     [77] " network record session extractor preprocess raw network"
     [78] "trace session contain messag follow will use"
     [79] " preprocess subsequ step analysi"
     [80] ""
     [81] " messag"
     [82] ""
     [83] ""
     [84] " preprocess messag x can model sequenc"
     [85] "byte x b b 255"
     [86] "infer common structur pool messag need similar"
     [87] "measur capabl focus analysi discrimin"
     [88] "featur account differ style like binari versus textual"
     [89] "protocol introduc two differ embed can"
     [90] "compress via statist test enabl focus analysi"
     [91] ""
     [92] " ngram"
     [93] ""
     [94] "one common approach domain natur languag process"
     [95] " map byte sequenc finitedimen featur space whose"
     [96] "dimens associ substr fix"
     [97] "length n formal can describ substr w bn"
     [98] " defin embed function b"
     [99] " follow"
    [100] ""
    [101] " w"
    [102] ""
    [103] ""
    [104] " simpli record whether specif ngram w occur "
    [105] "given string instanc 0"
    [106] ""
    [107] " 0t"
    [108] "n3 exampl can see correspond featur"
    [109] "space finit high dimension howev space"
    [110] "general spars popul allow effici"
    [111] "represent"
    [112] ""
    [113] " token"
    [114] ""
    [115] "anoth wellknown concept domain natur languag"
    [116] "process token byte sequenc via predefin"
    [117] "separ charact s embed b"
    [118] " map byte sequenc featur vector"
    [119] "record occurr possibl word w accord "
    [120] "separ w exampl"
    [121] " consid separ s"
    [122] "get follow embed ho let go 0"
    [123] ""
    [124] " 0t"
    [125] " similar ngram embed dimens"
    [126] " result featur space larg spars popul"
    [127] "therefor effici storag model also avail"
    [128] ""
    [129] " reduct"
    [130] ""
    [131] " find structur network communic analysi "
    [132] "focus featur discrimin messag "
    [133] "pool volatil featur like random generat nonc cooki"
    [134] "will occur lead unnecessari bloat vector"
    [135] "space hold true constant token protocol sinc"
    [136] " occurr messag will almost certain"
    [137] ""
    [138] "consequ use statist testdriven dimens reduct"
    [139] " allow us split featur space follow f"
    [140] "fconstant fvariabl fvolatil end"
    [141] "appli binomi see appendix featur"
    [142] "whether distribut frequenc approxim 1"
    [143] "correspond constant featur 0 volatil featur"
    [144] "respect applic multipl test correct"
    [145] " keep featur constant"
    [146] " volatil given statist signific level "
    [147] "005 simplifi featur space group togeth"
    [148] "featur exhibit correl near one"
    [149] ""
    [150] "given embed dimens reduct techniqu now"
    [151] "abl defin datadriven featur space messag allow"
    [152] "us introduc geometr concept like metric see"
    [153] "appendix open whole field"
    [154] "machin learn tool appli network communic "
    [155] "follow will assum total number featur f"
    [156] "w dimens reduct method keep just f"
    [157] " w featur embed function adjust"
    [158] "accord ie silent return just f reduc"
    [159] "featur"
    [160] ""
    [161] " event infer"
    [162] ""
    [163] ""
    [164] "messag occur specif event flow communic"
    [165] "often exhibit similar structur featur thus extract event"
    [166] "inform can exploit structur depend insid vector"
    [167] "space can defin metric captur notion similar"
    [168] " two messag instanc euclidean metric"
    [169] " y w"
    [170] "calcul distanc two point base occurr"
    [171] " w word contain corpus use metric"
    [172] "cluster algorithm can appli extract common structur "
    [173] " pool therebi indirect recov under event"
    [174] "inform"
    [175] ""
    [176] " infer structur network protocol trace suggest two"
    [177] "possibl cluster techniqu one protocol assembl"
    [178] " part one monolith communic token"
    [179] "weight accord absolut posit "
    [180] "messag obvious experiment free choos"
    [181] "appropri cluster techniqu hand found"
    [182] " two method work best protocol describ"
    [183] "kind"
    [184] ""
    [185] " cluster via matrix factor"
    [186] ""
    [187] " map network payload vector space induc geometri"
    [188] "reflect characterist captur reduc number featur"
    [189] " f instanc payload share sever substr appear"
    [190] "close wherea network payload differ content"
    [191] "exhibit larger geometr distanc vectori represent"
    [192] "network enabl us identifi semant compon"
    [193] " see appendix"
    [194] "particular appli concept matrix factor"
    [195] "identifi base direct vector space given "
    [196] "payload p pn first defin matrix"
    [197] "contain vector p column"
    [198] ""
    [199] " f n"
    [200] ""
    [201] ""
    [202] " determin compon seek represent "
    [203] "retain inform describ term base"
    [204] "direct can achiev term matrix factor"
    [205] " two matric b f e c"
    [206] " n e f"
    [207] ""
    [208] ""
    [209] " bc"
    [210] ""
    [211] " b1"
    [212] ""
    [213] ""
    [214] " c1 cn"
    [215] ""
    [216] ""
    [217] ""
    [218] " column b1 f b form new basi"
    [219] " n payload dimens base direct"
    [220] "bi associ featur w show"
    [221] " later experi relat base direct "
    [222] "featur can exploit construct templat matrix"
    [223] "factor column c1 c form"
    [224] "new coordin payload lowdimension"
    [225] "space coordin can use visual "
    [226] " lowdimension space"
    [227] ""
    [228] "anoth interest interpret matrix factor model"
    [229] " domain document cluster accord "
    [230] "follow document generat accord under"
    [231] "latent topic zj j e topic generat"
    [232] "specif word probabl pwizj specif document"
    [233] "assign topic probabl pzjd probabl"
    [234] " specif word occur document d can express"
    [235] "follow"
    [236] ""
    [237] "pwid"
    [238] ""
    [239] " matrix factor b c can also interpret "
    [240] "probabilist topic row b column c satisfi"
    [241] " axiom probabl see appendix"
    [242] ""
    [243] " general matrix factor method differ constraint"
    [244] "impos matric b c chapter studi two"
    [245] "standard techniqu wide use field statist "
    [246] "analysi princip compon analysi pca"
    [247] "nonneg matrix factor nmf"
    [248] ""
    [249] " compon analysi pca"
    [250] ""
    [251] " pca seek base direct orthogon captur"
    [252] "much varianc insid possibl formal ith"
    [253] "direct bi consecut maxim varianc bi"
    [254] " constraint base direct mutual"
    [255] "orthonorm"
    [256] ""
    [257] "bi b"
    [258] " b bj"
    [259] ""
    [260] ""
    [261] " set matrix factor correspond"
    [262] " singular valu decomposit l orthonorm basi"
    [263] "vector b equal first l leftsingular vector "
    [264] " coordin c correspond first rightsingular vector"
    [265] " multipli singular valu"
    [266] " pca entri b typic nonzero henc"
    [267] "featur contribut basi vector bi"
    [268] ""
    [269] " matrix factor"
    [270] ""
    [271] ""
    [272] "nonneg matrix factor nmf describ "
    [273] "approxim whole embed matrix fn contain"
    [274] "n point f reduc featur two strict"
    [275] "posit matric b fe c"
    [276] ""
    [277] " bc bc"
    [278] " c b c"
    [279] " bij 0 0"
    [280] ""
    [281] " inner dimens e matrix product b c chosen"
    [282] " e f lead even compact represent due"
    [283] " posit constraint matrix b can interpret new"
    [284] "basi messag matrix c contain"
    [285] " coordin newli span space "
    [286] " differ part coordin use ultim assign"
    [287] " messag cluster find posit maxim"
    [288] "weight shown nmf equival latent"
    [289] "semant index special kind topic"
    [290] " render nmf especi use context document"
    [291] "cluster"
    [292] ""
    [293] " sever possibl solut solv equat"
    [294] "see instanc lee99 hoyer04 heiler06 "
    [295] "stick practic implement introduc "
    [296] "base least squar approach altern"
    [297] "solv follow constraint least squar problem given"
    [298] "regular constant"
    [299] ""
    [300] ""
    [301] " b c"
    [302] " c b"
    [303] ""
    [304] ""
    [305] " correspond solut"
    [306] ""
    [307] ""
    [308] "c b b"
    [309] "b c c"
    [310] ""
    [311] ""
    [312] " regular constant can treat metaparamet "
    [313] "procedur choos crossvali sinc "
    [314] "number featur number sampl matrix can get"
    [315] "quit larg instanc ftp introduc later contain"
    [316] "rough 18 million sampl 90000 featur direct calcul"
    [317] "equat often infeas"
    [318] ""
    [319] "therefor devis reduc equival problem take"
    [320] "account dimens reduct step"
    [321] " duplic matrix"
    [322] " ie denot "
    [323] " f n matrix without duplic"
    [324] "column simplif equat note"
    [325] ""
    [326] ""
    [327] "ci b b ai"
    [328] ""
    [329] "henc can replac "
    [330] "equat obtain c duplic"
    [331] " result ci accord retriev c "
    [332] "simplif equat note"
    [333] ""
    [334] " c b c b 2"
    [335] ""
    [336] " w diagon matrix consist"
    [337] " number duplic correspond column"
    [338] " shown optim"
    [339] "problem equat right side"
    [340] "equat new object can solv"
    [341] ""
    [342] "b c w c c w"
    [343] ""
    [344] " two simplif allow us appli nmf even larg"
    [345] "set reduct accuraci appendix"
    [346] "describ replicateawar version nmf algorithm detail"
    [347] "togeth estim heurist inner dimens e"
    [348] " appli initi scheme"
    [349] ""
    [350] " cluster"
    [351] ""
    [352] " matrix factor techniqu good choic protocol"
    [353] " messag construct part protocol show"
    [354] "positiondepend featur sinc cluster step prisma"
    [355] "total independ concret algorithm use long "
    [356] "procedur assign cluster label messag experiment"
    [357] " fix matrix factor method free choos"
    [358] "appropri cluster tool choic take posit depend"
    [359] "featur account propos weight distanc measur"
    [360] ""
    [361] "dwx y w 101pw x 101pw y"
    [362] ""
    [363] " pw x return posit token w string x"
    [364] "distanc measur can use calcul distanc matrix d"
    [365] " subsequ form input singl linkag hierarch"
    [366] "cluster note can also restrict calcul "
    [367] "distanc matrix reduc matrix "
    [368] " save comput time also keep size d reason"
    [369] "rang"
    [370] ""
    [371] " statemachin"
    [372] ""
    [373] ""
    [374] "network communic driven under state machin"
    [375] " certain event trigger certain respons switch"
    [376] "proceed state sometim switch probabilist natur"
    [377] " instanc servic temporarili unavail can model"
    [378] " instanc login procedur 90 attempt"
    [379] "success"
    [380] ""
    [381] "one possibl way state machin network servic "
    [382] "probabilist way markov unobserv"
    [383] "state correspond intern logic state servic"
    [384] " messag sent network correspond emit"
    [385] "symbol use baumwelch algorithm enough"
    [386] " servic communic possibl estim"
    [387] "under hidden markov describ observ howev"
    [388] "baumwelch algorithm guarante found"
    [389] " highest likelihood global sens"
    [390] ""
    [391] " markov"
    [392] ""
    [393] "instead direct tri infer under hidden markov"
    [394] " start regular markov will later"
    [395] "simplifi minim hidden variant whole learn process"
    [396] "therefor determinist inher random like"
    [397] "initi matric baumwelch algorithm see"
    [398] "appendix detail approach"
    [399] "circumv problem find potenti nonoptim "
    [400] "determin come price well known fact hidden"
    [401] "markov model power regular one"
    [402] "summari trade potenti uncertainti decreas "
    [403] "complex therefor regular hypothes space"
    [404] ""
    [405] "given session inform preprocess step label"
    [406] "inform messag event cluster direct"
    [407] "learn regular markov event sequenc estim initi"
    [408] " transit probabl maximum likelihood"
    [409] "estim howev simpl markov drop"
    [410] "direct event ie event trigger client"
    [411] " server limit histori one messag due markov"
    [412] "assumpt ie generat next event depend just "
    [413] "previous one especi last limit strict"
    [414] "network communic sinc loos context "
    [415] "messag generat"
    [416] ""
    [417] " state space"
    [418] ""
    [419] " circumv limit regular markov "
    [420] "use convolut com version event"
    [421] "sequenc session follow"
    [422] ""
    [423] " event will annot reflect whether "
    [424] " generat client server side"
    [425] " horizon k convolut annot pad"
    [426] " event sequenc slide window size k record"
    [427] " occur ktupl"
    [428] ""
    [429] ""
    [430] " exampl assum observ event sequenc b c d"
    [431] " messag generat altern client"
    [432] "server horizon k 2 convolut event"
    [433] "sequenc "
    [434] " ac ac bs bs cc cc ds"
    [435] " new convolut event space e will contain 2e"
    [436] " 1k potenti event "
    [437] " start state calcul transit"
    [438] "probabl new convolut event space e"
    [439] " maximum likelihood estim specifi regular markov"
    [440] " annot event horizon k"
    [441] ""
    [442] " markov"
    [443] ""
    [444] " client server communic horizon least k2"
    [445] "necessari keep communic context involv"
    [446] "process even higher horizon might necessari lead "
    [447] "exponenti growth possibl state will see evalu"
    [448] " real network communic convolut state"
    [449] "space often spars popul yet result network can"
    [450] " larg make introspect human user difficult"
    [451] ""
    [452] " remedi propos follow minim algorithm boil"
    [453] " size markov preserv overal"
    [454] "capabl"
    [455] ""
    [456] " transform markov m determinist finit automaton"
    [457] " dfa m"
    [458] ""
    [459] " keep transit probabl bigger zero"
    [460] " associ state"
    [461] " transit dfa m accept new event"
    [462] " second state exampl transit connect state"
    [463] " ac bs state bs cc consum event cc"
    [464] ""
    [465] " appli dfa minim algorithm introduc"
    [466] " get equival dfa m "
    [467] " minim number state accept languag"
    [468] " side effect algorithm return assign"
    [469] " e m origin state "
    [470] " convolut event space e compress state"
    [471] " dfa m"
    [472] ""
    [473] ""
    [474] " result dfa m can use inspect "
    [475] "under state can interpret special hidden"
    [476] "markov instead observ convolut event"
    [477] "e now observ state m accord "
    [478] "assign e m found minim"
    [479] "algorithm metast subsum equival state will"
    [480] "therefor lead accept event sequenc "
    [481] "origin will show evalu "
    [482] "simplifi model drastic decreas size therefor"
    [483] "good candid analysi state machin human"
    [484] "administr"
    [485] ""
    [486] " templat rule"
    [487] ""
    [488] ""
    [489] " session can seen sequenc event trigger"
    [490] "specif state switch state machin learn general"
    [491] "inform flow process general messag"
    [492] "associ state templat consist fix"
    [493] "variabl part often fill content previous"
    [494] "messag method like matrix factor allow "
    [495] "extract static templat without take dynam inform"
    [496] "account session inform can use even extract fill"
    [497] "rule exploit learn markov now readi"
    [498] "give procedur reliabl extract templat rule "
    [499] "network servic hand"
    [500] ""
    [501] " templat"
    [502] ""
    [503] " event cluster step focus variabl yet neither constant"
    [504] " volatil featur identifi common pattern exchang"
    [505] "messag focus make sens identif"
    [506] "under event essenti featur back "
    [507] "generat valid protocolconform messag"
    [508] ""
    [509] ""
    [510] ""
    [511] ""
    [512] " state state b state c"
    [513] ""
    [514] " session 1 ftp 314 user anon 331 user anon ok"
    [515] " session 2 ftp 312 user ren 331 user ren ok"
    [516] ""
    [517] " session n ftp 20 user liz 331 user"
    [518] " liz ok"
    [519] ""
    [520] " templat ftp user 331 user"
    [521] " ok"
    [522] ""
    [523] ""
    [524] " templat generat simplifi ftp communic"
    [525] ""
    [526] ""
    [527] ""
    [528] " addit aspect extract generic messag templat"
    [529] " under state machin analyz servic "
    [530] "like exchang messag correl current state"
    [531] " servic thus valid assumpt assign"
    [532] "messag session accord state previous"
    [533] "extract state machin shown artifici exampl"
    [534] "figur look recur token "
    [535] "state generic templat can construct contain fix passag"
    [536] " variabl field accord distribut learn"
    [537] "pool"
    [538] ""
    [539] " detail templat infer procedur structur"
    [540] "follow"
    [541] ""
    [542] " token messag accord previous chosen embed"
    [543] " assign messag session state infer"
    [544] " markov"
    [545] " state markov"
    [546] ""
    [547] " analyz token messag assign state "
    [548] " number token"
    [549] " group assign messag number token process"
    [550] " group"
    [551] " messag group contain token "
    [552] " specif posit fix token record result"
    [553] " templat otherwis variabl field save"
    [554] ""
    [555] ""
    [556] " end procedur will templat state"
    [557] " markov repres generic messag might"
    [558] "occur note state might sever differ templat"
    [559] "assign accord observ length distribut ie"
    [560] "simplifi multipl align procedur extract"
    [561] "generic templat focus align messag "
    [562] "length"
    [563] ""
    [564] " rule"
    [565] ""
    [566] "find rule fill specif field templat"
    [567] "accord previous seen messag now amount simpl yet"
    [568] "power combin markov extract templat"
    [569] "session inform possibl combin templat"
    [570] "occurr horizon length k ie t1 t2 tk"
    [571] ""
    [572] " find messag assign k templat"
    [573] " occur session exact order"
    [574] " field f templat tk"
    [575] ""
    [576] " look rule fill f field content f"
    [577] " f templat t1 t2 tk f session"
    [578] " rule match just record token occur"
    [579] " train pool rule"
    [580] ""
    [581] ""
    [582] ""
    [583] " check rule describ tabl"
    [584] " procedur ensur inform found preced"
    [585] "messag can systemat reproduc content follow"
    [586] "messag f case will get copi instanc"
    [587] " exampl shown figur can observ"
    [588] " case field templat associ state c can"
    [589] "fill field previous messag rule act"
    [590] " fallback solut match found "
    [591] "pumpprim rule first messag session"
    [592] ""
    [593] ""
    [594] ""
    [595] ""
    [596] ""
    [597] " rule descript"
    [598] ""
    [599] " copi exact copi content one field anoth"
    [600] " seq copi numer field increment d"
    [601] " add copi content field add d front back"
    [602] " part copi front back part field split separ"
    [603] " s"
    [604] " fill field random pick previous seen d"
    [605] ""
    [606] ""
    [607] ""
    [608] " check build paramet"
    [609] " like d s automat infer "
    [610] " train"
    [611] ""
    [612] ""
    [613] ""
    [614] " network communic"
    [615] ""
    [616] " infer prisma now contain three part actual markov"
    [617] " infer templat rule set associ "
    [618] "templat use part prisma simul "
    [619] "communic devis live essenc network sensor len"
    [620] "depict algorithm addit infer"
    [621] " part modul initi role client server"
    [622] " simul note prisma role"
    [623] "agnost therefor can use simul side "
    [624] "communic allow us even let talk "
    [625] "run two instanc len differ role pass"
    [626] "messag generat one instanc vice versa"
    [627] ""
    [628] "appendix give complet exampl prisma"
    [629] "base simpl toy problem given network trace robot"
    [630] "communic environ behavior learn via prisma"
    [631] ""
    [632] ""
    [633] " live essenc network sensor len"
    [634] ""
    [635] ""
    [636] " templat rule role"
    [637] " activ"
    [638] " wait messag timeout t"
    [639] " m receiv"
    [640] " find match templat t accord "
    [641] " current state"
    [642] " split messag m accord t field"
    [643] " switch state state associ t"
    [644] ""
    [645] " random choos state s accord "
    [646] " transit probabl markovmodel"
    [647] " accord role"
    [648] " find rule accord previous"
    [649] " k horizon templat"
    [650] " appli rule fill new templat form"
    [651] " messag"
    [652] " send messag"
    [653] " current state s"
    [654] ""
    [655] ""
    [656] ""
    [657] ""
    [658] ""
    [659] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file004.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[2]]
    $content
     [1] " static content"
     [2] ""
     [3] ""
     [4] " present prisma method turn empir evalu"
     [5] " capabl differ secur applic evalu"
     [6] " first part process chain name embed"
     [7] "matrix factor step restrict evalu static"
     [8] "content apart give valuabl insight inner work"
     [9] " first step method show extract compon"
     [10] " matrix factor can use construct stateless"
     [11] "templat result base direct inclus"
     [12] "dynam inform form session inform enabl us"
     [13] " refin"
     [14] "templat accord occurr workflow "
     [15] "session evalu static content first studi"
     [16] "framework toy allow us establish"
     [17] "understand static templat infer communic"
     [18] "compar perform differ matrix factor method"
     [19] " proceed realworld"
     [20] "applic network trace contain malici communic"
     [21] " analyz static templat exploit vulner"
     [22] "attack sourc final"
     [23] "appli method context network anomali detect"
     [24] "limit process network reduc space give us"
     [25] " huge perform increas attain accuraci full"
     [26] " space"
     [27] ""
     [28] ""
     [29] " factor method"
     [30] ""
     [31] ""
     [32] " first experi consid artifici http"
     [33] "communic total control protocol syntax"
     [34] "semant simul web applic support three differ"
     [35] "type request whose network payload depict"
     [36] "figur first payload reflect request"
     [37] "static content second payload resembl search queri "
     [38] "last payload correspond administr request "
     [39] "action paramet one follow"
     [40] " request equip random"
     [41] "part name static web page search string "
     [42] "administr paramet simul usual fluctuat web"
     [43] "traffic"
     [44] ""
     [45] ""
     [46] ""
     [47] ""
     [48] " request static content"
     [49] " labelpositionbottomlin"
     [50] " get static3lpan6c2html http11"
     [51] " host wwwfoobarcom"
     [52] " accept"
     [53] ""
     [54] ""
     [55] " search queri"
     [56] " labelpositionbottomlin"
     [57] " get cgisearchphpseh0ykj3r3wd2i http11"
     [58] " host wwwfoobarcom"
     [59] " accept"
     [60] ""
     [61] ""
     [62] " administr request"
     [63] " labelpositionbottomlin"
     [64] " get cgiadminphpactionrenamepardbjh7hs0r5 http11"
     [65] " host wwwfoobarcom"
     [66] " accept"
     [67] ""
     [68] ""
     [69] " payload artifici "
     [70] ""
     [71] ""
     [72] ""
     [73] "use web applic generat 1000 network"
     [74] "payload uniform distribut three request type"
     [75] " appli first part prisma method detail"
     [76] " "
     [77] "use token basic string delimit select accord"
     [78] " specif http"
     [79] "d"
     [80] "base extract featur appli matrix factor"
     [81] "algorithm name princip compon analysi pca nonneg"
     [82] "matrix factor nmf determin base direct "
     [83] "vector space payload final construct static templat "
     [84] "base direct"
     [85] ""
     [86] " present extract featur tabl"
     [87] "featur consist 8 symbol correspond relev string"
     [88] " under web applic constant volatil token "
     [89] "filter token cooccur communic "
     [90] "coupl indic oper note featur"
     [91] " contain token relat http syntax therebi differ"
     [92] " previous approach reconstruct protocol grammar"
     [93] "specif"
     [94] ""
     [95] ""
     [96] ""
     [97] ""
     [98] ""
     [99] " symbol"
    [100] ""
    [101] " 1 5"
    [102] " 2 6"
    [103] " 3 s 7"
    [104] " 4 adminphp par 8"
    [105] ""
    [106] ""
    [107] " featur artifici string cooccur"
    [108] " network payload coupl use oper"
    [109] ""
    [110] ""
    [111] ""
    [112] "result applic matrix factor algorithm "
    [113] "artifici visual figur"
    [114] " algorithm pca nmf base direct matrix b shown"
    [115] " xaxi detail differ direct yaxi"
    [116] "contribut individu featur symbol"
    [117] ""
    [118] ""
    [119] ""
    [120] " 70 10 110cliptruewidth49"
    [121] " 70 10 110cliptruewidth49"
    [122] " base pca nmf artifici"
    [123] " color signifi intens entri rang"
    [124] " 1 red 1 blue"
    [125] ""
    [126] ""
    [127] ""
    [128] " techniqu perform matrix factor payload"
    [129] " matric differ signific pca yield posit"
    [130] "negat contribut matrix b indic differ"
    [131] "color although certain structur relat featur symbol"
    [132] "may deduc matrix clear separ differ"
    [133] "element possibl contrast nmf matrix show crisp"
    [134] "represent base direct static search request"
    [135] "clear reflect individu base direct remain base"
    [136] "direct correspond administr request differ"
    [137] "combin action type featur symbol "
    [138] "correct identifi"
    [139] ""
    [140] "due superior perform restrict analysi base"
    [141] "direct determin use nmf algorithm "
    [142] "follow static templat result sole posit entri "
    [143] "nmf matrix figur present"
    [144] "tabl templat accur captur"
    [145] "semant implement simpl web applic 7"
    [146] "templat construct cover static access web content"
    [147] "search queri differ administr task note two base"
    [148] "direct figur ident result "
    [149] "total 7 static templat templat even exhibit hierarch"
    [150] "structur templat 3 resembl basic administr request"
    [151] " follow templat special case particular"
    [152] "administr action"
    [153] ""
    [154] ""
    [155] ""
    [156] ""
    [157] ""
    [158] " static templat"
    [159] ""
    [160] " 1"
    [161] " 2 searchphp s"
    [162] " 3 action adminphp par"
    [163] " 4 action adminphp par move"
    [164] " 5 action adminphp par renam"
    [165] " 6 action adminphp par delet"
    [166] " 7 action adminphp par show"
    [167] ""
    [168] ""
    [169] " templat extract artifici "
    [170] " templat construct use token basic string nmf"
    [171] " matrix factor"
    [172] ""
    [173] ""
    [174] ""
    [175] " honeypot"
    [176] ""
    [177] ""
    [178] "network honeypot proven use instrument"
    [179] "identif analysi novel threat often howev"
    [180] "amount collect honeypot huge manual"
    [181] "inspect network payload becom tedious futil"
    [182] "propos prisma method allow analyz larg set"
    [183] "unknown traffic extract semant interest network featur"
    [184] "automat"
    [185] ""
    [186] " illustr util framework network collect"
    [187] "use webbas honeypot web"
    [188] " applic honeypot httpglastopforg honeypot captur"
    [189] "attack web applic remot file inclus rfi"
    [190] " sql inject attack expos typic pattern vulner"
    [191] "applic search engin honeypot deploy "
    [192] "period 2 month collect averag 3400 request per day"
    [193] " experi random pick 1000 request "
    [194] "collect appli framework use token under"
    [195] "featur particular extract 40 static templat use"
    [196] "base direct identifi nmf embed http payload"
    [197] "templat shown tabl"
    [198] " note 12"
    [199] " templat omit contain redund unspecif"
    [200] " inform"
    [201] ""
    [202] ""
    [203] ""
    [204] ""
    [205] ""
    [206] " static templat descript"
    [207] ""
    [208] " shellz csptxt semant rfi malwar"
    [209] " psybnc csptxt"
    [210] " botz bottxt"
    [211] " scannerz bottxt"
    [212] ""
    [213] " option http"
    [214] " vulner virtuemart"
    [215] " option itemid"
    [216] " show"
    [217] " exportphp phptxt"
    [218] " skin standard"
    [219] " vulner technot"
    [220] " smiley http"
    [221] " vulner gnuboard"
    [222] " skin http"
    [223] " file 1"
    [224] " file http"
    [225] " admin zefatxt"
    [226] " http fx29id1txt"
    [227] " appserv mainphp"
    [228] " vulner appserv"
    [229] " document media"
    [230] " vulner php"
    [231] " dir 1"
    [232] " misc rfi vulner"
    [233] " error phptxt bottxt"
    [234] " indexphp rawtxt"
    [235] " includ http"
    [236] ""
    [237] " com stealth"
    [238] " sourc attack"
    [239] " com compon"
    [240] " wwwhfsborg site 10225 img"
    [241] " eeng zefatxt"
    [242] " zerozoncokr photo count"
    [243] " musicadelibrerianet footer"
    [244] " forum smiley"
    [245] ""
    [246] " indexphp file miscellan templat"
    [247] " http path"
    [248] " path rawtxt"
    [249] " http zefatxt"
    [250] " option itemid"
    [251] " lib sourcedir"
    [252] " zero http"
    [253] " http id"
    [254] " http bjorktxt"
    [255] " media path"
    [256] " bottxt skin"
    [257] " http id"
    [258] ""
    [259] ""
    [260] " templat honeypot templat "
    [261] " construct use token basic string nmf matrix"
    [262] " factor"
    [263] ""
    [264] ""
    [265] ""
    [266] " extract static templat can classifi three categori"
    [267] " malwar "
    [268] " sourc exampl first templat reflect differ option"
    [269] "suppli webbas malwar malici function "
    [270] "prepar remot shell set irc bouncer"
    [271] " scan vulner host"
    [272] "clear manifest string templat follow"
    [273] "templat character vulner web applic includ"
    [274] "correspond file paramet name follow templat"
    [275] "correspond domain host name use sourc remot file"
    [276] "inclus often origin host also part "
    [277] "complet url discov final method also extract"
    [278] " miscellan direct map"
    [279] "vulner"
    [280] ""
    [281] "note although templat generat raw"
    [282] "http traffic syntact protocolspecif string "
    [283] "extract demonstr abil prisma focus semant"
    [284] "communic"
    [285] ""
    [286] " embed"
    [287] ""
    [288] ""
    [289] " show effici embed evalu capabl"
    [290] " prisma applic network intrus"
    [291] "detect techniqu anomali detect frequent appli"
    [292] "extens signaturebas intrus detect system "
    [293] "popular system sinc"
    [294] "enabl identif unknown novel network threat"
    [295] ""
    [296] " evalu intrus detect perform consid"
    [297] "three larger set network payload first "
    [298] "first08 contain http request monitor web server "
    [299] "research institut period 60 day second "
    [300] "blog09 contain request sever web blog run popular"
    [301] "platform span 33 day traffic third "
    [302] "ftp03 compris client side request ftp session record"
    [303] " 10 day lawrenc berkeley nation laboratori"
    [304] " addit benign inject network"
    [305] "attack traffic attack execut virtual"
    [306] "environ use common tool penetr test "
    [307] " care adapt match characterist "
    [308] " set see detail"
    [309] ""
    [310] " experi appli detect method similar work"
    [311] "wang et al centroid normal"
    [312] "network payload construct use ngram"
    [313] " use identifi unusu"
    [314] "network content addit consid second "
    [315] "ngram refin use reduc vector space formal"
    [316] "calcul matrix factor bc construct "
    [317] " follow b"
    [318] "ci calcul centroid lowerdimension space"
    [319] "obtain first 20 base direct nmf two model"
    [320] "train 1000 random drawn payload anomali"
    [321] "detect perform 200000 random chosen http request"
    [322] "20000 ftp session respect"
    [323] ""
    [324] "result shown receiv oper characterist roc curv"
    [325] " figur perform full reduc"
    [326] "centroid ident three set demonstr"
    [327] " base direct identifi nmf captur semant"
    [328] "inform under protocol suffici detect"
    [329] "anomali attack figur detail runtim"
    [330] "perform attain differ model reduc analysi"
    [331] "use lowerdimension space provid signific perform"
    [332] "gain regular anomali detect speedup factor 815"
    [333] "can observ clear indic util prisma"
    [334] "preprocess step anomali detect"
    [335] ""
    [336] ""
    [337] " curv"
    [338] " perform"
    [339] " curv runtim network anomali detect"
    [340] " templat construct use 4gram basic string"
    [341] " nmf matrix factor"
    [342] ""
    [343] ""
    [344] ""
    [345] ""
    [346] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file005.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[3]]
    $content
     [1] " dynam content"
     [2] ""
     [3] ""
     [4] " show prisma method benefit "
     [5] "incorpor dynam inform demonstr prisma capabl"
     [6] "learn also simul network communic real network"
     [7] "trace end use sever network trace record via"
     [8] " plug one part process pipelin"
     [9] " check qualiti accord remain"
     [10] " syntact semant featur simul"
     [11] "session ensur evalu prisma reallif"
     [12] "condit"
     [13] ""
     [14] " use raw test preprocess tool"
     [15] " effici featur dimens reduct"
     [16] " comparison heldout session assur"
     [17] " model mean learn model capabl replay"
     [18] " real session observ pool"
     [19] " check syntact semant featur simul"
     [20] " session guarante model communic"
     [21] " perspect"
     [22] ""
     [23] ""
     [24] " introduc set discuss"
     [25] " result featur space dimens reduct look"
     [26] " general properti learn pris model"
     [27] " complet correct "
     [28] "model conclud evalu "
     [29] " case studi malwar analysi show pris can use"
     [30] "applic domain beyond honeypot"
     [31] ""
     [32] " set dimens reduct"
     [33] ""
     [34] ""
     [35] " evalu prisma framework chosen three"
     [36] "repres set two textbas one"
     [37] "pure binari"
     [38] ""
     [39] " sip record real medium size telephoni"
     [40] " infrastructur contain rough 7 day communic 20"
     [41] " particip differ session initi protocol sip"
     [42] " client"
     [43] " dns domain name system dns request home network"
     [44] " 7 differ client collect one day heavi use"
     [45] " ftp file transfer protocol ftp "
     [46] " lawrenc berkeley nation laboratori contain"
     [47] " client server request 10 day communic"
     [48] ""
     [49] ""
     [50] "natur set vari size sip "
     [51] "mediums pool rough 35000 messag dns "
     [52] "contain just 6000 messag ftp compris near 18"
     [53] "million messag render biggest evalu"
     [54] " accommod differ properti set appli"
     [55] "differ embed sinc sip ftp consist humanread"
     [56] "text can token usual white space charact due"
     [57] " binari layout dns token approach"
     [58] " feasibl therefor chosen 2gram embed"
     [59] "dns set use 90 learn prisma"
     [60] " keep remain evalu carri "
     [61] ""
     [62] ""
     [63] " latex tabl generat r 2150 xtabl 156 packag"
     [64] " thu jun 14 143150 2012"
     [65] ""
     [66] ""
     [67] ""
     [68] ""
     [69] " size dimens kept"
     [70] " uniqu"
     [71] ""
     [72] " sip 34958 72937 039 258"
     [73] " dns 5539 6625 1315 3564"
     [74] " ftp 1760824 87140 217 024"
     [75] ""
     [76] ""
     [77] ""
     [78] " set give total number"
     [79] " messag number featur"
     [80] " dimens reduct step kept"
     [81] " uniqu give percentag featur messag"
     [82] " kept dimens reduct step"
     [83] ""
     [84] ""
     [85] ""
     [86] " result featur dimension reduct uniqu messag"
     [87] " shown tabl first thing note "
     [88] " dimens reduct step show excel behavior"
     [89] "relat number kept featur rang 04 sip 132"
     [90] " dns 22 ftp show extrem focus"
     [91] " eman dimension reduct direct consequ"
     [92] " relat number uniqu messag "
     [93] "rang 26 sip 356 dns 02"
     [94] "ftp strike differ dns sipftp term"
     [95] "reduct can clear explain differ conceptu layout"
     [96] " languag high compress binari format dns"
     [97] "protocol leav less room optim featur space"
     [98] "therefor also number uniqu messag dimens"
     [99] "reduct higher compar textbas protocol"
    [100] ""
    [101] "overal see dimens reduct high effect"
    [102] "even binari protocol focus vari part"
    [103] " messag uniqu messag reduc featur"
    [104] "space valuabl comput time can save render prisma"
    [105] "approach capabl model even big collect"
    [106] ""
    [107] " learn model"
    [108] ""
    [109] ""
    [110] "follow embed dimension reduct step appli"
    [111] "event cluster step describ "
    [112] " sip dns appli nmf cluster algorithm"
    [113] "sinc quick inspect show "
    [114] "partwholerelationship under nmf algorithm hold "
    [115] "two set relat short ftp messag follow less"
    [116] "fix setup render positiondepend cluster approach"
    [117] "better suit kind "
    [118] ""
    [119] ""
    [120] ""
    [121] ""
    [122] ""
    [123] " node coverag min dfa"
    [124] " coverag"
    [125] ""
    [126] " sip 148 145 100 98"
    [127] " dns 381 08 153 03"
    [128] " ftp 1305 08 653 04"
    [129] ""
    [130] ""
    [131] ""
    [132] " node prisma model "
    [133] " unoptim markov minim dfa relat"
    [134] " number potenti number node possibl"
    [135] ""
    [136] ""
    [137] ""
    [138] "tabl summar number node extract"
    [139] "markov relat number "
    [140] "potenti number node attain describ"
    [141] " see total number node"
    [142] " sip smallest yet relat coverag"
    [143] "highest dns ftp absolut number node higher"
    [144] " relat coverag potenti node space spars"
    [145] "indic inher depend relat coverag"
    [146] " estim number cluster applic dfa minim"
    [147] "algorithm markov signific reduc number"
    [148] "node model convert result network"
    [149] "dimens manag human user"
    [150] ""
    [151] ""
    [152] ""
    [153] ""
    [154] ""
    [155] " copi seq"
    [156] " add part "
    [157] " total"
    [158] ""
    [159] " sip 1916 77 135 52 1793 3972"
    [160] " dns 3142 4 0 0 3527 6673"
    [161] " ftp 532 18 253 35 4671 5509"
    [162] ""
    [163] ""
    [164] ""
    [165] " differ rule prisma model extract"
    [166] " differ set"
    [167] ""
    [168] ""
    [169] ""
    [170] " correspond number rule shown"
    [171] "tabl note ngram embed "
    [172] " rule deactiv sinc alreadi handl "
    [173] " rule see rule repres sip"
    [174] " exhibit higher number involv rule compar"
    [175] " set reflect high redund structur "
    [176] "protocol dns ftp inher variabl part server"
    [177] "name dns file name ftp result higher"
    [178] "number rule compar sip"
    [179] ""
    [180] " lal evaluationlensagainstlensdata20031009lbnltrain"
    [181] " savetransisitontru"
    [182] " s lalplay"
    [183] " zipslensgen sreplayedmsg"
    [184] " 11691"
    [185] " 1691495"
    [186] " 220 domain ftp server version wu2621 mon dec 30 165835 pst 2001 readi"
    [187] " 691495326 user anonym"
    [188] " 495326385"
    [189] " 331 guest login ok send complet email address password"
    [190] " 326385197 pass password"
    [191] " 38519720 230 guest login ok access restrict appli"
    [192] " 19720650 type"
    [193] " 20650423 200 type "
    [194] " 6504231660 pasv"
    [195] " 4231660179 227 enter passiv mode 1312431109240"
    [196] " 16601791627 retr groffperl11814i386rpm"
    [197] " 17916271525"
    [198] " 150 open binari mode connect groffperl11814i386rpm 56 byte"
    [199] " 16271525765"
    [200] ""
    [201] ""
    [202] ""
    [203] ""
    [204] " ftp server 2001 readi"
    [205] ""
    [206] " guest login ok send complet email address password"
    [207] " password"
    [208] " guest login access restrict appli"
    [209] ""
    [210] ""
    [211] ""
    [212] " enter mode"
    [213] ""
    [214] " open mode connect byte"
    [215] ""
    [216] " ftp session generat execut two prisma model"
    [217] " one client one server field"
    [218] " mark box exact copi rule fill gray"
    [219] ""
    [220] ""
    [221] ""
    [222] "figur give visual impress learn"
    [223] " ftp generat session simul"
    [224] "side communic prisma learn ftp"
    [225] " one execut act client "
    [226] "one act server see result log session"
    [227] " generat valid ftp start initi login"
    [228] "procedur client set communic"
    [229] "binari enter passiv mode get file "
    [230] "server note name file client request"
    [231] "copi correspond repli server show"
    [232] "power infer rule obvious byte size 56 "
    [233] "proper size request file sinc chosen random"
    [234] " rule messag valid ftp repli"
    [235] "show abil prisma even generat new messag seen"
    [236] " train pool"
    [237] ""
    [238] " correct"
    [239] ""
    [240] ""
    [241] " previous figur exampl show prisma"
    [242] "method produc relat condens model embed"
    [243] "space state machin question regard complet"
    [244] "correct model treat "
    [245] ""
    [246] ""
    [247] ""
    [248] " judg model take 10 "
    [249] "heldout simul either client server side"
    [250] "evalu whether learn contain path "
    [251] "generat session resembl sinc"
    [252] "transit probabilist ensur "
    [253] "path choos simul synchron actual"
    [254] "content session instanc session might contain"
    [255] "specif branch state machin occur just 5 "
    [256] "time like server overload error repli like allevi"
    [257] "probabilist effect repeat simul 100 time"
    [258] "introduc determin feed first two messag"
    [259] " session state first two"
    [260] "messag exchang align"
    [261] ""
    [262] " "
    [263] " shortcut use keep number repetit low "
    [264] " achiev effect just increment number"
    [265] " repetit unfortun render simul "
    [266] " ftp infeas due vast amount "
    [267] ""
    [268] " result simul report"
    [269] "figur use normal dameraulevenshtein"
    [270] "distanc similar 1 mean equal count"
    [271] "number insert delet substitut necessari"
    [272] "transform one string anoth posit session"
    [273] "take maximum attain similar repetit take"
    [274] "account probabilist effect describ"
    [275] ""
    [276] " normal done follow denot d"
    [277] " number insert delet substitut l"
    [278] " total length string similar "
    [279] " sl return 1 special case l equal 0"
    [280] ""
    [281] ""
    [282] ""
    [283] " "
    [284] ""
    [285] " "
    [286] ""
    [287] " "
    [288] ""
    [289] " maxim similar messag posit"
    [290] " session replay record normal edit distanc give"
    [291] " 10 equal messag size black bar"
    [292] " correspond frequenc equal messag size"
    [293] " dark gray bar similar rang 09 10"
    [294] " light gray bar similar rang 075 09"
    [295] ""
    [296] ""
    [297] ""
    [298] " sip observ number equal messag"
    [299] "rang 80 60 similar score almost never"
    [300] " 09 show learn model can correct remodel"
    [301] "holdout session dns behavior similar show"
    [302] "varianc due relat low number session 6"
    [303] "messag ftp show even better perform "
    [304] "prisma near messag show equal"
    [305] " posit six frequenc exact resembl stay alway"
    [306] " 70 show even complex protocol can accur"
    [307] "simul four step"
    [308] ""
    [309] ""
    [310] ""
    [311] "next focus syntact semant"
    [312] " generat messag syntact correct util"
    [313] "protocol filter network protocol analyz"
    [314] " ftp protocol check"
    [315] "valid command manual accord rfcs"
    [316] " check semant correct"
    [317] " appli follow rule"
    [318] ""
    [319] " sip messag session check whether"
    [320] " preserv"
    [321] " sinc tripl valu identifi sipsess"
    [322] " dns messag session repli check whether"
    [323] " queri session queri id"
    [324] " ftp ftp request check whether request"
    [325] " return repli code valid one accord rfcs"
    [326] ""
    [327] " session count number syntact semant"
    [328] "correct messag report relat frequenc correct messag"
    [329] " complet session addit session generat "
    [330] "complet evalu denot also"
    [331] "simul 100000 session side generat "
    [332] "denot "
    [333] ""
    [334] ""
    [335] ""
    [336] ""
    [337] ""
    [338] " syntax"
    [339] " semant"
    [340] ""
    [341] " unidir bidir unidir"
    [342] " bidir"
    [343] ""
    [344] " sip 1000 1000 0988 0945"
    [345] " dns 1000 1000 1000 0994"
    [346] " ftp 0999 0821 0934 0576"
    [347] ""
    [348] ""
    [349] ""
    [350] " session 100 syntact"
    [351] " semant correct messag"
    [352] " differ simul paradigm uni bidirect"
    [353] ""
    [354] ""
    [355] ""
    [356] " result shown tabl syntact"
    [357] "correct session almost alway perfect "
    [358] "bidirect simul ftp show relat"
    [359] "declin just 82 percent session total"
    [360] "correct regard semant dns show also near perfect"
    [361] "behavior perform sip 98 94"
    [362] " session total correct uni bidirect"
    [363] "simul respect also good rang "
    [364] "semant ftp unidirect case show good behavior"
    [365] " perform declin bidirect simul just 57"
    [366] " session total correct sinc ftp session tend "
    [367] "long investig correct detail"
    [368] "tabl split frequenc bin "
    [369] "observ bulk session 80 correct"
    [370] "messag combin higher length ftp"
    [371] "session show even difficult potenti vast"
    [372] "communic pattern prisma abl captur "
    [373] "syntax semant communic"
    [374] ""
    [375] ""
    [376] ""
    [377] ""
    [378] ""
    [379] " syntax"
    [380] " semant"
    [381] ""
    [382] " msgs correct unidir bidir"
    [383] " unidir bidir"
    [384] ""
    [385] " 100 0999 0821 0934 0576"
    [386] " 90 1000 0953 0988 0878"
    [387] " 80 1000 0996 1000 0982"
    [388] ""
    [389] ""
    [390] ""
    [391] " cumul syntact semant correct session"
    [392] " ftp"
    [393] ""
    [394] ""
    [395] ""
    [396] " summari evalu show infer pris model"
    [397] " compact show high degre complet well"
    [398] "syntact semant correct render model"
    [399] "readi deploy reallif network infrastructur act"
    [400] " honeypot specif design occur traffic "
    [401] "network contact honeypot can held high number"
    [402] " step gather indepth inform behavior intent"
    [403] " potenti intrud inform use"
    [404] "estim threat potenti infrastructur given time"
    [405] "point also learn attack mischief"
    [406] "conduct"
    [407] ""
    [408] " studi koobfac"
    [409] ""
    [410] ""
    [411] " appli prisma network traffic collect"
    [412] "malici softwar pick one specif class"
    [413] "malwar instanc use token embed partbas"
    [414] "cluster total 147 session 6674 messag"
    [415] "detail result depict"
    [416] "figur"
    [417] ""
    [418] ""
    [419] ""
    [420] ""
    [421] " state koobfac traffic upper part"
    [422] " correspond scan phase malwar"
    [423] " middl part handshak procedur infect machin"
    [424] " malwar get new list malwar final"
    [425] " download lower part state machin"
    [426] ""
    [427] ""
    [428] ""
    [429] " upper part see scan loop malwar"
    [430] "tri find infect server long server "
    [431] "answer specif format scan continu malwar"
    [432] " receiv correct repli state fs handshak procedur"
    [433] " malwar server take place follow "
    [434] "download cycl state ic malwar start download"
    [435] "first file list follow"
    [436] "state file download can nice seen"
    [437] " rule associ templat state kc"
    [438] "contain sever instanc follow path"
    [439] ""
    [440] "sysgetexetg14ex sysgetexems26ex"
    [441] "sysgetexehi15ex sysgetexebe18ex"
    [442] "sysgetexetw07ex sysgetexev2captchaex"
    [443] "sysgetexev2googlecheckex"
    [444] ""
    [445] " inspect extract statemachin associ templat"
    [446] " rule malwar analyst can gain insight inner work"
    [447] " malwar instanc collect network trace alon"
    [448] "render prisma valuabl tool beyond realm honeypot"
    [449] "applic"
    [450] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file006.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[4]]
    $content
     [1] " relat work"
     [2] ""
     [3] ""
     [4] " first discuss valuabl extens "
     [5] "practic deploy prisma model basic algorithm"
     [6] "describ produc accur model"
     [7] "reflect properti under pool administr"
     [8] "might want tune model better honeypot perform"
     [9] " compli privaci issu extens discuss"
     [10] " thorough discuss relat"
     [11] "work prisma framework given "
     [12] ""
     [13] " extens"
     [14] ""
     [15] ""
     [16] "deploy prisma model honeypot add practic"
     [17] "constraint hard algorithm fashion"
     [18] "instanc contain inform violat"
     [19] " privaci user system sinc clever intrud"
     [20] "tri infer kind via communic "
     [21] "honeypot use learn even contain"
     [22] "password secur relat inform privaci issu even"
     [23] "becom secur risk therefor devis extens"
     [24] " prisma method allow administr explor learn"
     [25] " adjust need part analysi framework"
     [26] "bridg gap statist real world"
     [27] "probabilist view transform back "
     [28] "easili comprehend also function view outsid world"
     [29] ""
     [30] " model field"
     [31] ""
     [32] " rule generat process describ"
     [33] " introduc rule "
     [34] "fallback solut case find rule type"
     [35] "associ specif field rule trigger"
     [36] "select random valu seen train"
     [37] "phase fill specif field certain valid"
     [38] " also sound approach afflict potenti privaci"
     [39] "problem correspond field contain privaci even secur"
     [40] "relat content like usernam password one definit"
     [41] "want leak kind inform therefor intellig"
     [42] " pragmat solut need field preserv privaci"
     [43] " secur issu"
     [44] ""
     [45] "one possibl solut problem infer perfield"
     [46] "languag instanc token observ content"
     [47] "contain rule ngram observ"
     [48] "string sequenc ngram generat can"
     [49] " leverag power markov chain "
     [50] "observ pool probabilist way now fill"
     [51] "field content just generat string accord"
     [52] " learn markov string will resembl observ"
     [53] "without direct return actual observ valu pool"
     [54] " suffici divers kind will privaci"
     [55] " secur preserv"
     [56] ""
     [57] ""
     [58] ""
     [59] "anoth issu deploy prisma honeypot real"
     [60] "world craft correct complet "
     [61] "servic also interest one potenti intrud might"
     [62] " traffic use learn contain flaw"
     [63] " allow normal valid input pass "
     [64] "instanc learn horizon 2 just observ"
     [65] "messag sequenc kind abe cde "
     [66] " handl messag input ade cbe"
     [67] "sinc observ behavior domain expert know"
     [68] " sequenc messag also valid obvious solut"
     [69] " problem just add miss edg markov"
     [70] ""
     [71] ""
     [72] "unfortun markov now capabl handl"
     [73] " unseen situat rule "
     [74] " handl case obvious yet extrem"
     [75] "power resort concept subrul instead just learn"
     [76] " rule horizon k also learn rule horizon"
     [77] "k1 k2 0 add len modul see"
     [78] "algorithm automat pick rule "
     [79] "highest possibl horizon carri simul instanc"
     [80] " case learn rule transit de"
     [81] " allow us go simul note"
     [82] "transit horizon k0 consist sole"
     [83] " rule allow generat messag even"
     [84] "total unobserv transit"
     [85] ""
     [86] " explor"
     [87] ""
     [88] ""
     [89] ""
     [90] ""
     [91] " explor allow play"
     [92] " observ behavior accord under"
     [93] " markov"
     [94] ""
     [95] ""
     [96] ""
     [97] " extract inform like content specif field "
     [98] " react certain input administr need power"
     [99] " explor tool simul communic show"
    [100] " under graph node graph select"
    [101] " associ templat rule inspect even edit"
    [102] ""
    [103] " extrem use reallif deploy program tool"
    [104] " sophist useabl gui timeconsum"
    [105] "task therefor decid develop proofofconcept"
    [106] "explor termin shown figur user"
    [107] "can play via command line interfac observ"
    [108] "control flow prerend graph markov albeit"
    [109] " simpl solut approach alreadi show good usabl"
    [110] "possibl thus full fledg explor gui"
    [111] "definit valuabl addit prisma framework"
    [112] ""
    [113] " templat match"
    [114] ""
    [115] "simul network traffic via len modul involv find"
    [116] "right templat see line algorithm"
    [117] " given input can solv just token"
    [118] "given input accord learn find onetoon"
    [119] "correspond base sequenc token possibl"
    [120] "templat pool approach might strict realworld"
    [121] "deploy henc intellig templat match algorithm"
    [122] " valuabl extens len modul"
    [123] ""
    [124] "one way approach problem exploit edit distanc"
    [125] " given token messag possibl templat find"
    [126] " onetoon match can order possibl templat accord"
    [127] " editdist input pick templat "
    [128] "minim one exploit inform found distanc matrix"
    [129] " dameraulevenshtein distanc two string match"
    [130] "omit token found result optim align "
    [131] "input templat"
    [132] ""
    [133] " approach"
    [134] ""
    [135] ""
    [136] "infer messag format under protocol"
    [137] "state machin relev problem today fast chang"
    [138] "network world also tackl differ direct"
    [139] " pure revers engin perspect open sourc"
    [140] "communiti tri fulli understand inner work"
    [141] "proprietari protocol order develop open implement"
    [142] "sever network servic eg smb icq skype work"
    [143] " done manual fashion special relev network"
    [144] "protocol analysi secur field led mani research"
    [145] "effort automat learn protocol state machin "
    [146] "format messag involv valid communic session"
    [147] ""
    [148] " work constitut first attempt extract"
    [149] " field protocol messag draw upon advanc"
    [150] "comput techniqu approach propos cluster"
    [151] "complet messag construct phylogenet tree order"
    [152] " guid process global sequenc align "
    [153] "needlemanwunsch algorithm roleplay"
    [154] "author build idea tackl problem automat"
    [155] "replay valid messag protocol although present"
    [156] "limit approach requir side communic"
    [157] "follow script use configur system"
    [158] "alreadi consid problem simul state depend"
    [159] "communic within scope replay"
    [160] " system present propos enhanc solut beyond heurist"
    [161] "introduc concept theorem prove weakest precondit"
    [162] "verif mean handl protocol depend"
    [163] ""
    [164] " similar approach specif secur applic also"
    [165] "focus replay valid messag introduc realm"
    [166] "honeypot scriptgen leita2006 low"
    [167] "interact honeypot learn simul communic pattern"
    [168] "vulner object scriptgen infer"
    [169] "accur specif protocol obtain maximum"
    [170] "inform exploit attempt servic although close"
    [171] "relat approach scriptgen design monitor"
    [172] "lowlevel attack implement wherea prisma enabl"
    [173] "collect track semant attack top "
    [174] "implement similar strain research"
    [175] "studi use token cluster individu messag"
    [176] " find field messag structur howev work infer"
    [177] " state machin protocol thus can use"
    [178] "simul network communic"
    [179] ""
    [180] "differ approach base dynam taint analysi "
    [181] "propos infer protocol specif"
    [182] " lin2008 wondracek2008cui2008 order overcom lack"
    [183] "semant cluster techniqu reli dynam binari"
    [184] "analysi network servic handl protocol"
    [185] "messag eas find keyword delimit unfortun"
    [186] " work defer task learn protocol state"
    [187] "machin extens work practic focus secur"
    [188] " carri devis dispatch"
    [189] "system capabl infiltr botnet whose oper may"
    [190] "base custom proprietari protocol abl "
    [191] " work messag side communic also"
    [192] " host level worth mention work "
    [193] " use binari analysi extract properti messag"
    [194] "memori buffer alreadi decrypt"
    [195] ""
    [196] "final build idea order"
    [197] "construct state machin protocol dynam"
    [198] "behavior applic implement protocol extent"
    [199] " work certain resembl nonetheless approach"
    [200] "free addit burden binari taint analysi sinc "
    [201] "fulli network base gather larg amount input trace"
    [202] " system therebi straightforward task"
    [203] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file007.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[5]]
    $content
     [1] " conclus"
     [2] ""
     [3] ""
     [4] " prisma present tool capabl learn"
     [5] "simul communic given servic network traffic"
     [6] "alon repres intern state machin servic"
     [7] " markov extract templat rule via align"
     [8] "collect session state machin prisma abl extract"
     [9] "inform necessari effici simul "
     [10] "pool evalu show viewpoint"
     [11] "complet syntact semant correct prisma"
     [12] "capabl emul reallif network traffic"
     [13] ""
     [14] " next goal deploy prisma honeypot dedic network"
     [15] "infrastructur evalu show prisma model"
     [16] "solid expect valuabl input reallif applic check"
     [17] " practic extens robustifi"
     [18] "approach addit state fuzz interest"
     [19] "applic prisma instanc one can use extract markov"
     [20] " find communic path insid state machin"
     [21] "occur seldom therefor tend rather untest"
     [22] "errorpron believ templat structur rule can"
     [23] "give valuabl clue field fuzz content"
     [24] "maxim probabl enforc error analysi"
     [25] "koobfac network trace prisma show method can readili"
     [26] "appli domain malwar analysi still refin"
     [27] " instanc find interest path state machin"
     [28] " malwar can enhanc usabl prisma scenario"
     [29] ""
     [30] " term theoret find show cours"
     [31] "dimension interfer model"
     [32] "approach obvious preprocess featur select"
     [33] " abl exploit fact resid much smaller"
     [34] "subspac high dimension featur space henc elimin"
     [35] "unnecessari redund featur can expos subspac even"
     [36] " use matrix factor method like replicateawar"
     [37] "nonneg matrix factor condens "
     [38] " bare distributionallik minimum allow us subsequ step"
     [39] " extract probabilist version abstract state machin"
     [40] "layer approach preprocess follow featur reduct follow"
     [41] " probabilist model also reflect softwar design"
     [42] "prisma layer implement tool chain consist"
     [43] "small special modul facilit debug"
     [44] " modul also allow later reus part "
     [45] "project"
     [46] ""
     [47] ""
     [48] ""
     [49] ""
     [50] ""
     [51] ""
     [52] ""
     [53] ""
     [54] " squar error mse "
     [55] " support vector regress model gaussian kernel repeat"
     [56] " train 50 train point vari blue line"
     [57] " denot mse train red line denot"
     [58] " mse independ high"
     [59] " complex ie low train error diverg even"
     [60] " real error due overfit effect"
     [61] ""
     [62] ""
     [63] ""
     [64] ""
     [65] " 225"
     [66] " 125"
     [67] " 025"
     [68] " blue line real"
     [69] " valu red line support vector regress model"
     [70] " gaussian kernel train 50 train point black dot "
     [71] " sinc see model high"
     [72] " complex ie low left plot exhibit overfit model"
     [73] " low complex ie high right plot clear"
     [74] " underfit "
     [75] ""
     [76] ""
     [77] ""
     [78] "detect unwant behavior network infrastructur often involv"
     [79] " rule base system monitor sensor observ"
     [80] "aberr prespecifi rule rule often"
     [81] "handcraft secur specialist analyz domain"
     [82] "build meaning model instanc seen"
     [83] "tabl specif malici request "
     [84] "malwar content botz bottxt therefor"
     [85] " domain expert build accord regular express "
     [86] "intrus prevent system constant monitor network traffic"
     [87] " search match regular express rule "
     [88] "payload"
     [89] ""
     [90] "obvious solut involv lot man power "
     [91] "method introduc previous chapter equip us power"
     [92] "tool support effort even better "
     [93] " rule automat given pool seen"
     [94] "chapter structur featur extract"
     [95] " joint similar metric open whole"
     [96] "toolbox machin learn transfer suitabl"
     [97] "vector space allow us pose rule learn problem"
     [98] "regress classif task"
     [99] ""
    [100] " instanc regress problem want learn predict"
    [101] "equat y fx consist point d1"
    [102] "x1 y1 yn assum"
    [103] " drawn iidfrom p one approach"
    [104] " solv problem minim mean squar error mse"
    [105] " given ie f denot estim "
    [106] "function"
    [107] ""
    [108] " f d1 yi fxi2"
    [109] ""
    [110] "basic without prior knowledg under generat"
    [111] "process infinit mani solut lead minim"
    [112] " mse train one trivial method just"
    [113] "memoiz valu train obvious"
    [114] "lead poor perform unseen therefor one often impos"
    [115] " structur restrict function class even introduc"
    [116] " regular function can direct control"
    [117] " socal paramet"
    [118] " instanc ridg regress impos linear function class"
    [119] "combin restrict result coeffici valu"
    [120] "shown 7hastie09 assum y fx"
    [121] " e 0 var"
    [122] " expect predict error err "
    [123] "regress fit fx input point x x use"
    [124] "squarederror loss"
    [125] ""
    [126] " errx ey fx2 x x"
    [127] ""
    [128] " fx fx2bias2 fx"
    [129] " fx e fx2var fx"
    [130] ""
    [131] ""
    [132] " total error can broken irreduc part"
    [133] "varianc squar bias estim far"
    [134] " real solut varianc estim"
    [135] "wigg estim meta paramet learn method"
    [136] "control complex higher complex less"
    [137] "bias higher varianc estim vice"
    [138] "versa therefor choos complex high can"
    [139] "perfect imit train bias tend zero will"
    [140] "suffer higher varianc unseen will"
    [141] "overfit contrari choos complex low"
    [142] "estim will exhibit high bias low varianc lead "
    [143] "underfit optim function class will lie"
    [144] " bias varianc balanc lead"
    [145] "overal smallest error estim learner"
    [146] ""
    [147] "figur show train error"
    [148] "evolv support vector regress model"
    [149] "gaussian kernel repeat train 50 train point vari"
    [150] "width gaussian kernel fix regular"
    [151] "paramet can observ train error "
    [152] "complex model ie lower high underestim "
    [153] "error independ furthermor can see "
    [154] " optim perform rough around"
    [155] " visual impress behavior "
    [156] "differ complex class shown"
    [157] "figur high complex "
    [158] "figur overfit train low complex"
    [159] " figur underfit real function optim"
    [160] " figur accur describ real function"
    [161] "class"
    [162] ""
    [163] "sinc reli train error determin"
    [164] "optim complex therefor find tool estim"
    [165] " error adjust complex via"
    [166] " meta paramet learn method crossvalid "
    [167] "defacto standard appli machin learn tune meta paramet"
    [168] " machin learn method supervis learn set see"
    [169] " stone74 geisser75 also"
    [170] " recent extens review method part "
    [171] "held back use get unbias estim"
    [172] " true general error crossvalid comput"
    [173] "quit demand though full grid search possibl"
    [174] "combin paramet candid quick take lot time even"
    [175] " one exploit obvious potenti parallel"
    [176] ""
    [177] "therefor crossvalid seldom execut full practic"
    [178] " differ heurist usual employ speed "
    [179] "comput exampl instead use full grid local search"
    [180] "heurist may use find local minima error see"
    [181] "instanc bergstra12 howev"
    [182] " general local search method guarante can"
    [183] "given qualiti found local minima anoth frequent"
    [184] "use heurist perform crossvalid subset "
    [185] " train full get accur"
    [186] "predict problem find right size "
    [187] "subset appar subset small reflect"
    [188] "true complex learn problem paramet select"
    [189] "crossvalid will lead underfit model hand"
    [190] " larg subset will take longer crossvalid finish"
    [191] ""
    [192] "appli kind heurist requir experienc"
    [193] "practition high familiar howev"
    [194] "effect play subset approach manag will"
    [195] "discuss depth given increas subset "
    [196] "minim error will converg often much earlier "
    [197] " error thus use subset systemat way open "
    [198] "promis way speed select process sinc train"
    [199] "model smaller subset much"
    [200] "timeeffici process care taken "
    [201] "increas avail sudden reveal structur "
    [202] " lead chang optim paramet"
    [203] "configur still will discuss depth "
    [204] "way guard chang point make heurist"
    [205] "take subset promis candid autom procedur"
    [206] ""
    [207] " chapter will propos method speed"
    [208] "crossvalid consid subset increas size"
    [209] "remov clear underperform paramet configur way"
    [210] " lead substanti save total comput time sketch"
    [211] " figur order account"
    [212] "possibl chang point sequenti test adapt"
    [213] " control zone rough speak certain number"
    [214] " allow failur paramet configur time"
    [215] " framework give statist guarante drop clear"
    [216] "underperform configur final add stop criterion"
    [217] " watch earli converg process speed "
    [218] "comput"
    [219] ""
    [220] " follow will first discuss effect take subset"
    [221] " learner crossvalid "
    [222] "present method fast crossvalid via sequenti test cvst"
    [223] " discuss theoret properti "
    [224] "method final evalu method"
    [225] "synthet realworld set"
    [226] " give"
    [227] "overview relat approach "
    [228] "conclud chapter impati practition may skip"
    [229] "theoret treatment focus selfcontain"
    [230] " describ cvst algorithm "
    [231] "evalu "
    [232] ""
    [233] ""
    [234] ""
    [235] ""
    [236] ""
    [237] ""
    [238] ""
    [239] ""
    [240] " time consumpt 5fold crossvalid cv"
    [241] " left fast crossvalid via sequenti test cvst"
    [242] " right cv calcul configur"
    [243] " full cvst algorithm use increas"
    [244] " subset drop signific underperform"
    [245] " configur step result drastic decreas"
    [246] " total calcul time"
    [247] ""
    [248] ""
    [249] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file008.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[6]]
    $content
     [1] " subset"
     [2] ""
     [3] ""
     [4] " approach base take subset speed"
     [5] "crossvalid main question therefor whether can"
     [6] "reliabl estim best paramet configur alreadi"
     [7] "subset mild condit will prove general"
     [8] "converg estim also discuss can expect"
     [9] "estim much better practic theoret result"
     [10] "develop"
     [11] ""
     [12] " sake simplic will consid follow slight"
     [13] "simplifi formal crossvalid procedur"
     [14] "replac empir error expect risk "
     [15] "therebi also remov repetit involv typic run"
     [16] "kfold crossvalid way deal "
     [17] "addit inaccuraci involv estim error "
     [18] "finit make argument precis"
     [19] " actual compar error enc defin"
     [20] " equat anoth empir error"
     [21] " emc defin independ sampl instead compar"
     [22] " enc ec expect error paramet"
     [23] " configur c technic mean "
     [24] " consid enc emc enc ec emc"
     [25] " ec get essenti result addit"
     [26] " limit process m consid limit "
     [27] " size also tend infin sake simplic"
     [28] " drop detail"
     [29] ""
     [30] "assum train given inputoutput pair xi"
     [31] "yi drawn iidfrom probabl distribut"
     [32] "px y usual also assum "
     [33] " loss function given "
     [34] " overal error expect risk predictor g"
     [35] " given rg e y x y px y"
     [36] ""
     [37] " possibl paramet configur c let gnc"
     [38] " predictor learn paramet c c first n"
     [39] "train exampl crossvalid basic tri identifi"
     [40] "best paramet c given train size n minim"
     [41] " expect risk normal done comput best"
     [42] "paramet configur term empir risk holdout"
     [43] " present discuss assum estim"
     [44] " suffici accur expect risk"
     [45] ""
     [46] " now also consid differ train size "
     [47] "effect deal task minim sequenc function"
     [48] "e c defin"
     [49] ""
     [50] ""
     [51] " enc rgnc"
     [52] ""
     [53] ""
     [54] " thus interest minimum subset size k"
     [55] "ek relat en question link "
     [56] "asymptot behavior en minimum en"
     [57] "converg will ek en eventu"
     [58] ""
     [59] "now general expect en converg "
     [60] "exampl might paramet encod way"
     [61] " need scale sampl size ie might"
     [62] " valu c sampl size n correspond fnc "
     [63] "function fn complex set condit"
     [64] "paramet configur c might work well subset size k"
     [65] "becom suboptim choic larger sampl size"
     [66] ""
     [67] " will therefor assum fix c"
     [68] "enc exist use standard techniqu straightforward"
     [69] " prove follow result see appendix"
     [70] " proof"
     [71] ""
     [72] ""
     [73] ""
     [74] " let c finit assum fix c enc"
     [75] " ec probabl follow hold"
     [76] ""
     [77] " converg c n"
     [78] ""
     [79] " cenc ec 0"
     [80] ""
     [81] " probabl"
     [82] " minimum let cn "
     [83] " encn c enc c ec"
     [84] " c ec"
     [85] ""
     [86] " ecn ec 2 c enc ec"
     [87] ""
     [88] " moreov"
     [89] ""
     [90] " ecn ec 0 probabl"
     [91] ""
     [92] " subset "
     [93] " 0 exist number n n n k"
     [94] " n k n"
     [95] ""
     [96] " p encn"
     [97] ""
     [98] ""
     [99] ""
    [100] ""
    [101] " sake simplic assum c finit"
    [102] "result can like extend continu paramet space"
    [103] "signific technic overhead"
    [104] ""
    [105] "theorem basic prove asymptot"
    [106] "can expect get good estim right choic paramet"
    [107] "train size n subset k result assum paramet"
    [108] "configur encod way independ size"
    [109] " train hing uniform converg "
    [110] "possibl paramet choic"
    [111] ""
    [112] "now well result describ practic find"
    [113] "figur show error typic"
    [114] "exampl train support vector regress svr subset "
    [115] "full train consist 500 point "
    [116] " sinc introduc "
    [117] "paramet kernel width gaussian kernel use"
    [118] " regular paramet valu shown alreadi"
    [119] "optim regular paramet sake"
    [120] "simplic"
    [121] ""
    [122] " see minimum converg rather quick first plateau"
    [123] " 15 03 approxim toward"
    [124] "lower one 25 17 also optim one train"
    [125] " size n 500 see uniform converg main"
    [126] "drive forc fact error small kernel width still"
    [127] " far apart even minimum alreadi converg"
    [128] ""
    [129] " follow help continu discuss within"
    [130] "empir risk minim framework assum learner"
    [131] "train pick minim empir risk"
    [132] " hypothesi h set one can write differ"
    [133] " expect risk learn predictor rgn "
    [134] "bay risk r follow see also section121"
    [135] " section243"
    [136] ""
    [137] " rgn r"
    [138] ""
    [139] " error"
    [140] " r"
    [141] " error"
    [142] ""
    [143] " estim error measur far chosen one"
    [144] " asymptot optim approxim error"
    [145] "measur differ best possibl "
    [146] "hypothesi class true function"
    [147] ""
    [148] ""
    [149] ""
    [150] ""
    [151] ""
    [152] " error svr"
    [153] " sinc introduc"
    [154] " can observ shift optim"
    [155] " gaussian kernel finegrain structur "
    [156] " problem seen enough figur b approxim"
    [157] " error indic black solid line estim error"
    [158] " black dash line asymptot"
    [159] " approxim error plot blue dash line one can see"
    [160] " uniform approxim estim error main"
    [161] " drive forc instead decay approxim error"
    [162] " smaller kernel width togeth increas estim"
    [163] " error small kernel width make sure minimum converg quick"
    [164] ""
    [165] ""
    [166] "use decomposit can interpret figur follow see"
    [167] "figur kernel width"
    [168] "basic control approxim error "
    [169] "18 result hypothesi class coars repres"
    [170] "function consider becom smaller reach"
    [171] "level bay risk indic dash blue"
    [172] "line even larger train size can assum will"
    [173] "stay level even smaller kernel size"
    [174] ""
    [175] " differ blue line upper line show"
    [176] "estim error estim error extens studi"
    [177] " statist learn theori known link differ"
    [178] "notion complex like vcdimens"
    [179] "fatshatt dimens norm "
    [180] "reproduc kernel hilbert space rkhs"
    [181] "typic result show estim error can bound term"
    [182] " form"
    [183] ""
    [184] " rgn rn o nn"
    [185] ""
    [186] " dh notion complex under hypothesi"
    [187] "class figur mean can expect estim"
    [188] "error becom larger smaller kernel width"
    [189] ""
    [190] " basic order paramet configur accord"
    [191] " complex make three observ"
    [192] ""
    [193] " paramet small complex larg kernel"
    [194] " width approxim error will high estim"
    [195] " error will small"
    [196] " paramet high complex approxim error"
    [197] " will small even optim estim error will larg"
    [198] " also see figur"
    [199] " approxim error seem decreas faster increas"
    [200] " complex estim error increas"
    [201] ""
    [202] " combin estim smaller train size tend"
    [203] "underestim true complex approxim"
    [204] "error quick decreas minimum also converg true"
    [205] "one fact estim error larger complex"
    [206] "model act guard choos complex model"
    [207] ""
    [208] "unfortun exist theoret result abl bound"
    [209] "error suffici tight make argument"
    [210] "exact particular speed converg minimum"
    [211] "hing tight lower bound approxim error "
    [212] "realist upper bound estim error approxim error"
    [213] " studi exampl paper "
    [214] " paper prove upper bound "
    [215] "rate also worstcas rate like close enough"
    [216] " true error hand mechan lead fast"
    [217] "converg minimum plausibl look concret"
    [218] "exampl therefor will assum follow"
    [219] " locat best paramet configur might initi"
    [220] "chang becom less stabl quick will use"
    [221] "sequenti test introduc zone ensur"
    [222] " method robust initi chang"
    [223] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file009.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[7]]
    $content
     [1] " valid via sequenti test cvst"
     [2] ""
     [3] ""
     [4] " consid usual supervis learn set "
     [5] "consist point d1 x1 y1 yn"
     [6] " assum drawn iidfrom"
     [7] "px y learn algorithm"
     [8] " depend sever paramet c goal select"
     [9] "paramet c learn predictor g best"
     [10] "general error respect loss function"
     [11] " full kfold"
     [12] "crossvalid estim best paramet split "
     [13] " k part use k1 part train estim"
     [14] "error remain part"
     [15] ""
     [16] " approach attempt speed process take subsampl"
     [17] "size sn 1 s start "
     [18] "full paramet candid elimin clear"
     [19] "underperform candid step execut main"
     [20] "loop algorithm page"
     [21] "perform follow main part given subset "
     [22] ""
     [23] ""
     [24] " procedur transform pointwis error "
     [25] " remain configur binari top flop scheme"
     [26] " line"
     [27] " drop signific loser configur along way"
     [28] " line use test sequenti analysi"
     [29] " framework"
     [30] " appli robust distribut free test techniqu allow"
     [31] " earli stop procedur seen enough"
     [32] " stabl paramet estim line"
     [33] ""
     [34] ""
     [35] " follow will discuss individu step "
     [36] "algorithm conceptu overview one iter procedur"
     [37] "depict figur refer"
     [38] ""
     [39] " step cvst shown situat"
     [40] " step s 10 point"
     [41] " learn configur c1 ck error"
     [42] " calcul current d1 dn"
     [43] " transform binari perform indic trace"
     [44] " configur filter via sequenti analysi ck1"
     [45] " ck drop procedur check whether"
     [46] " remain configur perform equal well past stop"
     [47] " case see appendix "
     [48] " complet exampl run"
     [49] ""
     [50] " transform error"
     [51] ""
     [52] " robust transform perform configur "
     [53] "binari inform whether among topperform"
     [54] "configur turn flop reli distributionfre"
     [55] "test basic idea calcul pointwis perform "
     [56] "given configur point use modelbuild"
     [57] "process group togeth best perform configur show"
     [58] "similar behavior"
     [59] ""
     [60] "suppos situat depict figur"
     [61] " k remain configur c1 c2 ck"
     [62] " order accord mean perform now want find"
     [63] " smallest index t configur c1 c2"
     [64] " ct show behavior point d1 d2"
     [65] " dn main idea robust test rank"
     [66] "configur ci 1 t accord individu"
     [67] "perform jth point result rank ri"
     [68] " j h0 configur act similar sum "
     [69] "rank point st rt "
     [70] "similar t 1 t case regress task"
     [71] "individu perform measur realvalu residu "
     [72] "procedur implement sketch"
     [73] " classif task observ binari perform "
     [74] "procedur can appli instead see"
     [75] "appendix summari test find"
     [76] "smallest index t appli increas subset "
     [77] "configur point t1 show signific"
     [78] "effect given signific level result flag"
     [79] " configur 1 t top configur "
     [80] "remain t1 k configur flop configur note"
     [81] " increment procedur multipl test situat"
     [82] "sinc interest joint infer hypothes"
     [83] " use individu decid whether can observ"
     [84] "signific effect due addit new configur"
     [85] ""
     [86] " actual calcul error function"
     [87] " calcperform appli increment build process"
     [88] "ie ad step line increas"
     [89] "train pool allow onlinealgorithm adapt"
     [90] " also increment error calcul "
     [91] "remain point avail crossvalid"
     [92] " result first step collect trace matrix see"
     [93] "figur top right show gradual"
     [94] "transform last 10 step procedur highlight"
     [95] "result last"
     [96] ""
     [97] " signific loser"
     [98] ""
     [99] ""
    [100] " transform error scaleindepend top flop"
    [101] "scheme can now whether given paramet configur "
    [102] "overal loser sequenti test binari random variabl"
    [103] "address analysi framework develop"
    [104] " origin appli context"
    [105] "product qualiti assess compar two product process"
    [106] "biolog set stop bioassay soon gather"
    [107] "lead signific result"
    [108] ""
    [109] " main idea follow one observ sequenc"
    [110] "iidbinari random variabl z1 z2 want"
    [111] " whether variabl distribut accord h0"
    [112] " h1 signific level "
    [113] "accept h1 h0 can control via"
    [114] "metaparamet comput"
    [115] "likelihood far observ reject one "
    [116] "hypothesi respect likelihood ratio larger "
    [117] "factor control metaparamet can shown "
    [118] "procedur intuit geometr represent depict"
    [119] "figur lower left binari observ"
    [120] "record cumul sum time step sum exceed"
    [121] "upper red line accept h1 sum lower red"
    [122] "line accept h0 sum stay two red line"
    [123] " draw anoth sampl"
    [124] ""
    [125] "wald requir choos sinc"
    [126] " main goal use sequenti elimin"
    [127] "underperform choos paramet "
    [128] " h1 configur win postpon long"
    [129] "possibl will allow cvst algorithm keep configur"
    [130] " evid perform definit show "
    [131] " overal loser configur time want"
    [132] "maxim area configur elimin region denot"
    [133] " loser fig reject mani loser"
    [134] "configur way possibl"
    [135] ""
    [136] ""
    [137] " sa 1"
    [138] ""
    [139] ""
    [140] " defin "
    [141] "triangular area accept h0 sa"
    [142] " earliest step accept h1 variabl"
    [143] " defin total number step"
    [144] ""
    [145] "use result global optim"
    [146] "equat can solv follow"
    [147] ""
    [148] " 05"
    [149] " 10"
    [150] ""
    [151] ""
    [152] " averag sampl number"
    [153] " expect number step given will yield"
    [154] "decis real 10 note sequenti analysi"
    [155] "formal requir iidvari might true"
    [156] "configur transform winner configur later"
    [157] "therebi chang behavior flop "
    [158] "topconfigur therefor tune procedur use"
    [159] "sequenti analysi framework just decis whether"
    [160] "configur overal loser adjust"
    [161] " switch role keep potenti configur long"
    [162] "possibl just drop trace statist signific"
    [163] "correspond binomi 05 detail "
    [164] "open sequenti analysi pleas consult see"
    [165] "instanc general overview sequenti"
    [166] "test procedur"
    [167] ""
    [168] ""
    [169] ""
    [170] " main loop"
    [171] ""
    [172] ""
    [173] " conf"
    [174] " gettest"
    [175] " tracesc s performancec s 0"
    [176] " activec true"
    [177] " 1 "
    [178] " pointwiseperf"
    [179] " calcperformancedata activ"
    [180] " performanceact s averageperformancepointwiseperf"
    [181] " trace topconfspointwiseperf s 1"
    [182] " activ flopconfstest tracesact 1s fals"
    [183] " similarperformancetracesact s"
    [184] " break"
    [185] ""
    [186] ""
    [187] ""
    [188] " selectwinnnerperform activ"
    [189] ""
    [190] ""
    [191] ""
    [192] ""
    [193] " stop final winner"
    [194] ""
    [195] ""
    [196] "final employ earli stop rule line"
    [197] "take last column trace matrix"
    [198] "check whether remain configur perform equal well"
    [199] " past use cochran q "
    [200] " similarperform procedur figur"
    [201] "illustr complet run cvst algorithm rough 600"
    [202] "configur configur mark red correspond "
    [203] "flop configur black one top"
    [204] "configur final configur mark gray "
    [205] "drop via sequenti cvst algorithm small"
    [206] "zoomin lower part pictur show last"
    [207] " remain configur step "
    [208] "use earli stop criterion can see procedur"
    [209] "keep go heterogen behavior remain"
    [210] "configur mix redblack configur perform"
    [211] "equal well past near black earli stop "
    [212] " see signific effect anymor procedur stop"
    [213] ""
    [214] " upper plot show run "
    [215] " cvst algorithm rough 600 configur step"
    [216] " configur mark top black flop red drop"
    [217] " gray blowup show situat step 5 7 without"
    [218] " drop entri earli stop rule take affect step 7"
    [219] " remain configur perform equal well step 5 7"
    [220] ""
    [221] "final procedur selectwinn line"
    [222] " win configur pick configur "
    [223] "surviv step follow remain configur"
    [224] "determin rank step accord averag perform"
    [225] " step averag rank last"
    [226] " step pick configur lowest"
    [227] "mean rank way make use accumul"
    [228] " cours procedur restrict view last"
    [229] " observ also take account "
    [230] "optim paramet might chang increas size sinc"
    [231] "focus recent observ biggest model"
    [232] "alway pick configur suitabl "
    [233] "size hand"
    [234] ""
    [235] " cvst"
    [236] ""
    [237] ""
    [238] " cvst algorithm number metaparamet "
    [239] "experiment determin beforehand give"
    [240] "suggest choos paramet paramet"
    [241] "control signific level step similar"
    [242] "behavior suggest usual level "
    [243] "005 furthermor control signific"
    [244] "level h0 configur loser h1 configur"
    [245] " winner respect suggest asymmetr setup set"
    [246] " 01 sinc want drop loser configur relat"
    [247] "fast 001 sinc want realli sure "
    [248] "accept configur overal winner final "
    [249] " 3 6 20 "
    [250] " observ choic work well practic"
    [251] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file010.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[8]]
    $content
     [1] " properti cvst algorithm"
     [2] ""
     [3] ""
     [4] " introduc overal concept cvst algorithm"
     [5] "now focus theoret properti ensur proper"
     [6] "work procedur exploit guarante under"
     [7] "sequenti test framework show experiment can"
     [8] "control procedur work stabl regim furthermor"
     [9] "prove error bound cvst algorithm addit show"
     [10] " cvst algorithm can use work best given time budget"
     [11] ""
     [12] " bound stabl regim"
     [13] ""
     [14] ""
     [15] " discuss perform "
     [16] "configur might chang feed learn algorithm"
     [17] " therefor reason algorithm exploit learn"
     [18] "subset must capabl deal difficulti"
     [19] " potenti chang point behavior certain"
     [20] "configur investig theoret"
     [21] "properti cvst algorithm make particular suitabl"
     [22] " learn increas subset "
     [23] ""
     [24] " first properti open sequenti employ cvst"
     [25] "algorithm come handi control overal converg process"
     [26] " assur configur drop prematur"
     [27] ""
     [28] " zone given cvst"
     [29] " algorithm signific level "
     [30] " top flop configur respect maxim number step"
     [31] " s global win configur loos first"
     [32] " cp iter long "
     [33] ""
     [34] " 0 sz "
     [35] " 2"
     [36] ""
     [37] " probabl configur drop cvst"
     [38] "algorithm zero"
     [39] ""
     [40] ""
     [41] ""
     [42] " detail proof defer appendix"
     [43] ""
     [44] ""
     [45] " consequ lemma "
     [46] "experiment can direct control via signific level"
     [47] " iter prematur drop"
     [48] " occur therefor guid whole process stabl"
     [49] "regim configur will see enough show"
     [50] "real perform"
     [51] ""
     [52] "equip properti can now take thorough look "
     [53] "worst case perform cvst algorithm suppos global win"
     [54] "configur constant mark loser secur"
     [55] "zone amount avail point "
     [56] "suffici show superior configur given"
     [57] " global win configur now see enough mark"
     [58] " win configur binar process throughout"
     [59] "next step probabl can give exact error bound"
     [60] " overal process solv specif recurr"
     [61] ""
     [62] " worstcas"
     [63] " scenario error probabl cvst algorithm global"
     [64] " winner configur label constant loser "
     [65] " secur zone reach can calcul probabl"
     [66] " configur endur sequenti recurr"
     [67] " scheme count number remain path end "
     [68] " nonlos regiontrim30pt 40pt 30pt 40ptcliptru"
     [69] ""
     [70] "figur give visual impress "
     [71] "worst case analysi exampl 20 step cvst execut"
     [72] " win configur generat straight line zero "
     [73] "secur zone 7 approach bound error fast"
     [74] "crossvalid now consist essenti calcul probabl"
     [75] "mass end nonlos region"
     [76] " follow lemma show can express number path"
     [77] " lead specif point graph twodimension"
     [78] "recurr relat"
     [79] ""
     [80] " relat"
     [81] " denot pr c number path lead point"
     [82] " rc lie lower decis boundari l0 "
     [83] " sequenti given worst case scenario describ "
     [84] " number path can calcul follow"
     [85] ""
     [86] "pr c"
     [87] ""
     [88] " 1 r0 c sz"
     [89] " 1 r c sz 1"
     [90] " pr c1 pr1 c1 l0c r c sz 1"
     [91] " 0"
     [92] ""
     [93] ""
     [94] ""
     [95] ""
     [96] ""
     [97] " split proof 4 case"
     [98] ""
     [99] " first case definit configur "
    [100] " straight line zero secur zone sz"
    [101] " second case describ diagon path start point"
    [102] " 1 sz 1 construct path 1 mean diagon 0"
    [103] " mean one step right diagon path can just reach"
    [104] " singl combin name straight line one"
    [105] " third case actual recurr given point"
    [106] " lower decis bound l0 number path"
    [107] " lead point equal number path lie"
    [108] " direct left point plus path lie direct"
    [109] " diagon downward point first path"
    [110] " point can reach direct step right "
    [111] " latter current point can reach diagon step"
    [112] " upward sinc option "
    [113] " construct equal hold"
    [114] " last case describ path either lie"
    [115] " lower decis bound therefor end loser region"
    [116] " diagon can thus never reach"
    [117] ""
    [118] ""
    [119] ""
    [120] " recurr visual figur"
    [121] "number grid give number valid nonlos path"
    [122] "can reach specif point recurr now abl"
    [123] "prove global worstcas error probabl fast"
    [124] "crossvalid"
    [125] ""
    [126] " bound cvst suppos global"
    [127] " win configur reach secur zone constant"
    [128] " loser trace switch winner configur "
    [129] " success probabl error cvst"
    [130] " algorithm erron drop configur can determin"
    [131] " follow"
    [132] ""
    [133] " p 1 l0 1r"
    [134] " pi"
    [135] " r"
    [136] ""
    [137] ""
    [138] ""
    [139] " basic idea use number path lead nonlos"
    [140] "region calcul probabl configur actual"
    [141] "surviv correspond last column exampl"
    [142] "figur sinc outcom "
    [143] "binar process binomi variabl success"
    [144] "probabl first diagon path probabl"
    [145] " next path probabl"
    [146] " last viabl path"
    [147] "reach point l0 1 "
    [148] "complet probabl surviv configur sum"
    [149] " correspond number path lemma sinc "
    [150] "interest complementari event subtract result sum"
    [151] " one conclud proof"
    [152] ""
    [153] ""
    [154] "note earli stop rule interfer bound"
    [155] " worst case inde process goe maxim"
    [156] "number step sinc probabl mass will"
    [157] "maxim spread due linear lower decis boundari "
    [158] "correspond expon maxim earli stop rule"
    [159] "termin process reach maximum number step"
    [160] " result error probabl will lower given bound"
    [161] ""
    [162] " bound fast"
    [163] " crossvalid proven theorem differ"
    [164] " success probabl maxim step size mark"
    [165] " global trend fit loess curv given dot line "
    [166] " datatrim10pt 12pt 10pt 12ptcliptru"
    [167] ""
    [168] " error bound differ success probabl propos"
    [169] "sequenti depict"
    [170] "figur first can observ relat"
    [171] "fast converg overal error increas maxim number"
    [172] " step impact error margin shown"
    [173] "success probabl iefor instanc 095 error"
    [174] "near converg optimum 005 therefor chosen"
    [175] "scheme allow us control secur zone also"
    [176] " small impact overal error probabl "
    [177] "show practic sequenti ratio fast"
    [178] "crossvalid procedur use statist test can"
    [179] "balanc need conserv retent configur"
    [180] "long possibl statist control drop"
    [181] "signific loser configur near impact overal"
    [182] "error probabl analysi assum experiment"
    [183] "chosen right secur zone learn problem hand"
    [184] "small size happen secur zone chosen"
    [185] " small therefor chang point global win"
    [186] "configur might lie outsid secur zone will"
    [187] "occur often today size set analyz behavior"
    [188] "cvst circumst appendix give"
    [189] "complet view properti algorithm"
    [190] ""
    [191] " valid time budget"
    [192] ""
    [193] " time"
    [194] " consumpt cubic learner step calcul"
    [195] " subset calcul time tf"
    [196] " full adjust accord s"
    [197] " step process assum drop r k remain"
    [198] " configur"
    [199] ""
    [200] " cvst algorithm can use box speed"
    [201] "regular crossvalid aforement properti "
    [202] "procedur come handi face situat optim"
    [203] "paramet configur found given fix comput"
    [204] "budget time suffici perform full"
    [205] "crossvalid amount process "
    [206] "big explor suffici space paramet grid ordinari"
    [207] "crossvalid reason time cvst algorithm allow"
    [208] "get select inform given"
    [209] "specifi constraint"
    [210] ""
    [211] "basic achiev calcul maxim step paramet"
    [212] " lead near coverag avail time budget t"
    [213] "depict figur given k paramet"
    [214] "configur prespecifi secur zone bound s"
    [215] " 0 s 1 ensur configur"
    [216] "drop prematur comput demand cvst algorithm"
    [217] " approxim sum time need step"
    [218] "involv calcul k configur "
    [219] "step r k configur 0 r 1"
    [220] " will see experiment evalu assumpt"
    [221] " given drop rate 1r lead form time consumpt"
    [222] " depict figur quit"
    [223] "common observ drop rate correspond overal difficulti"
    [224] " problem hand"
    [225] ""
    [226] "given comput time tf need perform calcul"
    [227] " full prove appendix"
    [228] " optim maximum step paramet cubic learner can calcul"
    [229] "follow"
    [230] ""
    [231] " tf k1rs3 tf r k1rs4rtf"
    [232] " k k1rs3tf r k 2t1rs4rtf"
    [233] " k"
    [234] ""
    [235] " calcul maxim number step given time"
    [236] "budget t can use result lemma"
    [237] "determin maxim given fix yield"
    [238] " request secur zone bound"
    [239] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file011.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[9]]
    $content
     [1] ""
     [2] ""
     [3] ""
     [4] " evalu cvst algorithm real investig"
     [5] "perform control set regress"
     [6] "classif task introduc special tailor set"
     [7] "highlight overal behavior stress fast"
     [8] "crossvalid procedur evalu choic learn"
     [9] "method influenc perform cvst algorithm compar"
     [10] "kernel logist regress klr vector"
     [11] "machin svm classif problem kernel ridg regress"
     [12] "krr versus regress problem use gaussian"
     [13] "kernel scholkopf2000 experi"
     [14] "use 10 step cvst paramet set"
     [15] "describ give us upper bound "
     [16] "expect speed gain note get even higher speed gain"
     [17] " either lower number step increas "
     [18] "practic point view believ set studi"
     [19] "high realist"
     [20] ""
     [21] " set"
     [22] ""
     [23] " assess qualiti cvst algorithm first examin"
     [24] "behavior control set seen motiv"
     [25] " specif learn problem might sever layer"
     [26] "structur can reveal learner enough "
     [27] "avail instanc figur can see"
     [28] " first optim plateau occur real"
     [29] "optim paramet center around thus real optim"
     [30] "choic just becom appar seen 200"
     [31] "point"
     [32] ""
     [33] " construct learn problem regress"
     [34] " classif task pose sever problem cvst"
     [35] "algorithm stop earli will return suboptim"
     [36] "paramet evalu differ intrins dimension"
     [37] " various nois level affect perform "
     [38] "procedur classif task use sine"
     [39] " consist sine uniform sampl rang"
     [40] "control intrins dimension d"
     [41] ""
     [42] "y"
     [43] " n2 x 0 2 d"
     [44] " n 05 d 50 100"
     [45] " label sampl point just sign y"
     [46] "regress task devis sinc "
     [47] "consist sinc function overlay highfrequ sine"
     [48] "y x d x5"
     [49] " n2 x"
     [50] " n 02 d 3 4"
     [51] ""
     [52] " set generat 1000 point run 10"
     [53] "step cvst compar result normal"
     [54] "10fold crossvalid full record "
     [55] "error addit 10000 point time consum "
     [56] "paramet search explor paramet grid contain 610 equal"
     [57] "space paramet configur method"
     [58] " 29 3 01 05"
     [59] " svmsvr 6 2"
     [60] "klrkrr respect process repeat 50 time gather"
     [61] "suffici interpret overal process"
     [62] ""
     [63] ""
     [64] " mean squar error left relat speed gain right "
     [65] " sine "
     [66] ""
     [67] " result sine can seen"
     [68] "figur upper boxplot show"
     [69] "distribut differ mean squar error best"
     [70] "paramet determin cvst normal crossvalid low"
     [71] "nois set n 025 cvst algorithm find optim"
     [72] "paramet normal crossvalid intrins"
     [73] "dimension d50 d100 cvst algorithm get stuck"
     [74] " suboptim paramet configur yield increas"
     [75] "classif error compar normal crossvalid"
     [76] "tendenc slight increas high nois set n 05"
     [77] "yield broader distribut classif method use seem"
     [78] " direct influenc differ svm klr show"
     [79] "near similar behavior pictur chang look "
     [80] "speed gain svm near alway rang 15 19"
     [81] "klr show speedup 20 60 time varianc "
     [82] "speed gain general higher compar svm seem "
     [83] "direct consequ inner work klr basic main"
     [84] "loop perform step matrix invers whole kernel"
     [85] "matrix calcul coeffici converg obvious"
     [86] "converg criterion lead relat widespread distribut"
     [87] " speed gain compar svm perform"
     [88] ""
     [89] " configur"
     [90] " step sine "
     [91] ""
     [92] "figur show distribut "
     [93] "number remain configur step cvst"
     [94] "algorithm low nois set upper row can observ"
     [95] "tendenc bigger drop rate d100 high nois"
     [96] "set lower row observ steadi increas kept"
     [97] "configur combin higher spread "
     [98] "distribut overal see effect drop rate"
     [99] "configur set svm klr show near similar"
    [100] "behavior higher speed gain klr seen "
    [101] "direct consequ algorithm influenc"
    [102] " cvst algorithm"
    [103] ""
    [104] ""
    [105] " mean squar error left plot relat speed gain right"
    [106] " plot sinc "
    [107] ""
    [108] " perform sinc shown"
    [109] "figur first strike"
    [110] "observ transit cvst algorithm can"
    [111] "observ intrins dimension d4 point"
    [112] "overal excel perform cvst algorithm verg"
    [113] "choos suboptim paramet configur behavior "
    [114] "evid high nois set svr near alway show"
    [115] "smaller differ krr capabl delay declin"
    [116] " high nois set least part speed gain observ"
    [117] "near constant differ dimension nois level"
    [118] " rang 15 20 svr 60 80 krr"
    [119] ""
    [120] " configur"
    [121] " step sinc "
    [122] ""
    [123] " direct consequ behavior can observ"
    [124] " number remain configur shown"
    [125] "figur compar classif"
    [126] "experi drop much drastic intrins"
    [127] "dimension nois level show small influenc higher"
    [128] "dimension nois level yield remain configur"
    [129] " overal varianc distribut much smaller "
    [130] "classif experi"
    [131] ""
    [132] ""
    [133] " mean squar error svmsvr increas size"
    [134] " sine left sinc right set ad"
    [135] " cvst algorithm converg correct paramet"
    [136] " configur"
    [137] ""
    [138] " figur examin"
    [139] "influenc perform cvst algorithm"
    [140] " sine sinc abl"
    [141] " estim correct paramet configur nois"
    [142] "dimension set feed cvst enough"
    [143] " limit experi "
    [144] " svmsvr method sinc full crossvalid klrkrr"
    [145] " taken much time comput clear cvst capabl"
    [146] "extract right paramet configur increas"
    [147] "amount 2000 5000 point render method"
    [148] "even suitabl big scenario abund cvst"
    [149] "will abl estim correct paramet much smaller time"
    [150] "frame"
    [151] ""
    [152] " set"
    [153] ""
    [154] " demonstr overal perform cvst algorithm"
    [155] "control set will investig perform real life"
    [156] " well known benchmark set classif pick"
    [157] "repres choic set ida benchmark repositori"
    [158] "see"
    [159] "httpwwwmldataorg furthermor ad first two class"
    [160] " entri "
    [161] " follow procedur"
    [162] " paper sampl 2000 point class "
    [163] "learn estim error remain point"
    [164] "regress pick use add"
    [165] " delv"
    [166] "repositori httpwwwcstorontoedu"
    [167] ""
    [168] " process follow first scale "
    [169] "case regress also scale depend variabl split"
    [170] " half use one part train "
    [171] " estim error process repeat 50 time"
    [172] "get suffici statist perform method "
    [173] " artifici set compar differ error"
    [174] " speed gain fast normal crossvalid "
    [175] "paramet grid 610 valu "
    [176] " adjust"
    [177] "rang 59 0"
    [178] " adjust small structur found set"
    [179] ""
    [180] ""
    [181] " mean squar error upper plot relat speed gain lower"
    [182] " plot set"
    [183] ""
    [184] "figur show result "
    [185] "classif set left side regress set"
    [186] "right side upper panel depict differ mean squar"
    [187] "error mse classif task differ never"
    [188] "exceed two percent point show although fast"
    [189] "crossvalid procedur case seem pick suboptim"
    [190] "paramet impact fals decis small "
    [191] "hold true regress task sinc depend variabl"
    [192] " problem standard differ mse valu"
    [193] "compar observ classif task see"
    [194] "just small differ mse although problem"
    [195] "cvst algorithm pick suboptim paramet even "
    [196] "differ error alway relat small learner"
    [197] "hard impact behavior just "
    [198] " see signific differ "
    [199] "correspond method term speed gain see much"
    [200] "divers vari pictur overal speed improv klr"
    [201] " krr higher svm svr reach 120 time"
    [202] "compar normal crossvalid regress task general seem"
    [203] " solv faster classif task can clear"
    [204] "explain look trace"
    [205] "figur classif task number"
    [206] " kept configur general much higher "
    [207] "regress task furthermor can observ sever type"
    [208] "difficulti learn problem instanc "
    [209] " seem much difficult "
    [210] " also reflect differ"
    [211] "speed improv seen previous figur"
    [212] ""
    [213] ""
    [214] " configur step set"
    [215] ""
    [216] " summari evalu benchmark set show "
    [217] "cvst algorithm give huge speed improv compar normal"
    [218] "crossvalid see nonoptim choic"
    [219] "configur total impact error never except"
    [220] "high keep mind chosen paramet"
    [221] " cvst algorithm give impress maxim attain"
    [222] "speedup conserv set trade comput time"
    [223] "lower impact error"
    [224] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file012.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[10]]
    $content
     [1] " relat work"
     [2] ""
     [3] ""
     [4] "sequenti test use extens machin learn"
     [5] "context summar work relat"
     [6] " cvst algorithm discuss variant"
     [7] " sequenti evalu perform cvst"
     [8] "algorithm shown socal sequenti"
     [9] " lack essenti properti open variant wald use"
     [10] " cvst algorithm underlin optim "
     [11] "open wald learn increas subset "
     [12] ""
     [13] " test machin learn"
     [14] ""
     [15] "use statist test sequenti analysi framework order"
     [16] " speed learn topic sever line"
     [17] "research howev exist bodi work most focus"
     [18] "reduc number evalu focus overal"
     [19] "process elimin candid best "
     [20] "knowledg new concept can appar combin"
     [21] " alreadi avail race techniqu reduc total"
     [22] "calcul time"
     [23] ""
     [24] " introduc socal"
     [25] " race base nonparametr hoeffd bound "
     [26] "mean error step algorithm new point"
     [27] " evalu remain model confid interv"
     [28] " error updat accord model whose confid"
     [29] "interv error lie outsid least one interv "
     [30] "better perform drop devis"
     [31] " similar rang algorithm use concept pac learn game"
     [32] "theori differ hypothes order expect util"
     [33] "accord algorithm seen far "
     [34] "hoeffd race emphasi approach lie reduc"
     [35] "number evalu"
     [36] ""
     [37] " concept race extend "
     [38] "introduc upper bound learner loss function "
     [39] "exampl procedur allow earli stop learn"
     [40] "process loss near optim infinit"
     [41] " appli race domain evolutionari"
     [42] "algorithm extend framework use friedman "
     [43] "filter nonpromis configur use"
     [44] "similar concept context boost filterboost"
     [45] " introduc empir bernstein bound extend"
     [46] " filterboost framework race algorithm "
     [47] "case bound use estim error within specif"
     [48] " region given probabl use"
     [49] "concept sequenti test speed boost process"
     [50] "control number featur evalu "
     [51] "sampl similar fashion approach use"
     [52] " increas speed evalu "
     [53] "perceptron speed pegaso"
     [54] "algorithm use partial leaveoneout evalu"
     [55] " perform get estim overal perform"
     [56] " use pick probabl best race"
     [57] "concept appli wide varieti domain like reinforc"
     [58] "learn timet"
     [59] " show relev pratic impact topic"
     [60] ""
     [61] " first sight multiarm bandit problem"
     [62] " also seem relat problem"
     [63] " anoth way multiarm bandit problem number"
     [64] "distribut given task identifi distribut"
     [65] " largest mean chosen sequenc sampl "
     [66] "individu distribut round agent choos one"
     [67] "distribut sampl typic find balanc"
     [68] " explor differ distribut reject distribut"
     [69] " seem promis focus candid get"
     [70] " accur sampl"
     [71] ""
     [72] " look similar set also wish identifi"
     [73] "promis candid reject underperform configur"
     [74] "earli process main differ "
     [75] "multiarm bandit set assum distribut fix"
     [76] "wherea specif deal distribut chang"
     [77] " sampl size increas lead introduct "
     [78] "safeti zone among thing therefor multiarm bandit"
     [79] "set applic across differ sampl size"
     [80] ""
     [81] " hand multiarm bandit approach possibl"
     [82] "extens speed comput within fix train size"
     [83] "either test similar hoeffd race alreadi mention"
     [84] " elimin comput individu fold "
     [85] "crossvalid procedur underperform configur"
     [86] ""
     [87] " versus close sequenti test"
     [88] ""
     [89] " alreadi introduc sequenti"
     [90] "test pioneer monitor likelihood"
     [91] "ratio"
     [92] ""
     [93] ""
     [94] " hi x 1"
     [95] ""
     [96] ""
     [97] "hypothes h1 accept contrari h0"
     [98] "accept b neither condit appli"
     [99] "procedur accept either two hypothes need"
    [100] " b chosen error probabl "
    [101] "two decis exceed "
    [102] "respect proven open sequenti"
    [103] "probabl ratio wald optim sens compar"
    [104] " test power requir averag fewest"
    [105] "observ decis test scheme wald call"
    [106] " sinc procedur potenti go forev"
    [107] "long leav abtunnel"
    [108] ""
    [109] ""
    [110] " open design wald procedur led develop "
    [111] "differ kind sequenti test number observ"
    [112] " fix beforehand"
    [113] " instanc"
    [114] " clinic studi might imposs ethic prohibit"
    [115] "use potenti go forev unfortun none"
    [116] " socal close exhibit optim criterion"
    [117] "therefor choos one least simul studi show"
    [118] " best behavior term averag sampl number statist"
    [119] "method base gambler ruin scenario"
    [120] " player fix fortun decid play n"
    [121] "game fn fa fb probabl player"
    [122] "fortun fa stake b will ruin oppon fortun fb"
    [123] " exact n game follow recurr hold"
    [124] ""
    [125] "fn fa fb"
    [126] ""
    [127] " 0 fa 0 n 0 fb 0"
    [128] " 1 n 0 fa 0 fb 0"
    [129] " fn1 fa1 fbb"
    [130] " 1 fn1 fab fbb"
    [131] ""
    [132] ""
    [133] " step player can either win game probabl"
    [134] "win 1 oppon lose stake b "
    [135] "player now given n x y game player won y"
    [136] " player b won x game will stop either "
    [137] "follow condit hold"
    [138] ""
    [139] "y bx fa y"
    [140] " y bx fb y"
    [141] ""
    [142] ""
    [143] " formul cast gambler ruin problem waldlik"
    [144] "scheme just observ cumul win player "
    [145] "check whether reach lower upper line now choos"
    [146] "fa fb fn 05 fa fb"
    [147] "construct allow us check whether given"
    [148] "configur perform wors 05 ie cross lower"
    [149] "line can therefor flag overal loser control"
    [150] "error probabl see detail"
    [151] " close design spicer pleas consult"
    [152] ""
    [153] "sinc simul studi show close variant "
    [154] "sequenti test exhibit low averag sampl number statist"
    [155] "first look runtim perform cvst algorithm"
    [156] "equip either open close sequenti "
    [157] "influenti paramet term runtim "
    [158] "paramet principl larger number step lead robust"
    [159] "estim also increas comput time studi"
    [160] "effect differ choic paramet simul "
    [161] "sake simplic assum binari top flop scheme"
    [162] "consist independ bernoulli variabl "
    [163] " 09 10 00 01 "
    [164] " open close sequenti compar relat"
    [165] "speedup cvst algorithm compar full 10fold"
    [166] "crossvalid case learner cubic"
    [167] ""
    [168] " speed gain fast"
    [169] " crossvalid compar full crossvalid assum"
    [170] " train time cubic number sampl shown simul"
    [171] " runtim 10fold crossvalid differ problem class"
    [172] " differ loserwinn ratio easi 31 medium 11 hard 13"
    [173] " 200 resampl"
    [174] ""
    [175] "figur show result simul"
    [176] "runtim differ set overal speedup much higher"
    [177] " close sequenti indic aggress behavior"
    [178] "compar conserv open altern test show"
    [179] " highest increas rang 10 20 step rapid"
    [180] "declin toward higher step number term speed close"
    [181] "sequenti definit beat conserv open"
    [182] ""
    [183] ""
    [184] ""
    [185] " negat generat close top open bottom"
    [186] " sequenti nonstationari configur ie "
    [187] " given chang point bernoulli variabl chang"
    [188] " indic valu 10"
    [189] ""
    [190] "figur reveal"
    [191] " speed gain come price apart control "
    [192] "secur zone number fals drop configur much"
    [193] "higher open sequenti definit advantag"
    [194] " open term speed fals negat rate close"
    [195] " render useless cvst algorithm"
    [196] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file013.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[11]]
    $content
     [1] " conclus"
     [2] ""
     [3] ""
     [4] " present method speed crossvalid procedur"
     [5] "start subset full train size identifi"
     [6] "clear underperform paramet configur earli focus"
     [7] " promis candid larger subset size "
     [8] "discuss take subset theoret"
     [9] "advantag compar heurist like local search "
     [10] "paramet effect error systemat"
     [11] " can understood statist one hand show"
     [12] " optim configur converg true one sampl size"
     [13] "tend infin also discuss concret set "
     [14] "differ behavior estim error approxim error lead"
     [15] " much faster converg practic insight led "
     [16] "introduct safeti zone sequenti test"
     [17] "ensur underperform configur remov"
     [18] "prematur minima converg yet experi"
     [19] "show procedur lead speed 120 time"
     [20] "compar full crossvalid without signific increas"
     [21] " predict error"
     [22] ""
     [23] " will interest combin method procedur"
     [24] "like hoeff race algorithm multiarm bandit"
     [25] "problem furthermor get accur converg bound even"
     [26] "finit sampl size set anoth topic futur research"
     [27] " moment cvst algorithm precondit "
     [28] "optim paramet learn method independ "
     [29] "train size standard machin learn"
     [30] "algorithm adher precondit method like"
     [31] "knearest neighbor exhibit connect optim"
     [32] "paramet train size valuabl extens"
     [33] " cvst incorpor kind knowledg algorithm"
     [34] " lower precondit learner make fast"
     [35] "crossvalid procedur avail learn method"
     [36] ""
     [37] " structur point view can see problem"
     [38] "select optim learner given problem can deconstruct"
     [39] " decoupl sever layer first defin problem"
     [40] "domain complex solv via expect risk "
     [41] "learner exploit converg risk subset abl"
     [42] " repres perform learner binari fashion whether"
     [43] " belong top perform round observ"
     [44] "properti increas subset size can give statist bound"
     [45] "guarante overal perform via sequenti test"
     [46] " even use robust test earli stop whole procedur"
     [47] " case chang can expect layer"
     [48] "probabilist preprocess model allow stepbystep"
     [49] "construct process combin one toolchain solv"
     [50] "problem select plan releas toolbox"
     [51] " open sourc public avail cran packag see"
     [52] " cvst algorithm can use one layer "
     [53] "project"
     [54] ""
     [55] ""
     [56] ""
     [57] ""
     [58] ""
     [59] ""
     [60] ""
     [61] ""
     [62] " network infrastructur intern network"
     [63] " separ socal demilitar zone dmz dmz"
     [64] " contain server access outsid "
     [65] " secur firewal web server harden "
     [66] " applic firewal act revers proxi outsid"
     [67] ""
     [68] ""
     [69] ""
     [70] "respons secur incid requir accur analysi"
     [71] " under detect incid "
     [72] "network secur two kind model establish"
     [73] " secur defin via rule kind"
     [74] " traffic will allow wherea secur"
     [75] " fix allow traffic therefor case "
     [76] " secur intrus detect system issu"
     [77] "alarm one rule match incom traffic case"
     [78] " secur alarm issu "
     [79] "observ traffic deviat allow traffic"
     [80] ""
     [81] "apart choos type secur network"
     [82] "administr also choos suitabl point deploy"
     [83] " network infrastructur figur show"
     [84] "simpl network infrastructur intern network "
     [85] "client organ resid socal demilitar zone"
     [86] "dmz servic access"
     [87] " outsid organ entri dmz seal "
     [88] "firewal act central point entri departur"
     [89] "network traffic intern network dmz "
     [90] "outsid world dmz act anoth safeti zone protect"
     [91] "client organ servic insid dmz "
     [92] "restrict access intern network thus one "
     [93] "servic get compromis attack will abl spread insid"
     [94] " intern network"
     [95] ""
     [96] " harden specif servic one can also deploy applic"
     [97] "firewal like revers proxi shown"
     [98] "figur servic act like protect"
     [99] "web server ie request web server will first"
    [100] "process revers proxi socal web applic"
    [101] "firewal waf inspect incom request applic level"
    [102] "sinc waf know request web server"
    [103] "encod hypertext transfer protocol http detect"
    [104] "malici behavior can take place finegrain level"
    [105] ""
    [106] " intrus detect system mere report suspici"
    [107] "network traffic intrus system ip react"
    [108] "real time suspici activ either"
    [109] "traffic part request promin exampl"
    [110] " system inlin deploy just"
    [111] "like firewal figur waf"
    [112] " act like revers proxi figur"
    [113] "solut work rulebas detect engin defin"
    [114] "negat secur ie unless rule insid"
    [115] "system specif exploit these system abl protect"
    [116] " infrastructur servic hand"
    [117] ""
    [118] "approach base negat secur model prone"
    [119] " two major drawback first requir extens applic"
    [120] "attack knowledg maintain reliabl rule base furthermor"
    [121] " capabl detect known attack leav system wide"
    [122] "open socal attack pattern"
    [123] " avail suffic rule base simpli uptod"
    [124] " order attack slip thus focus posit"
    [125] "secur learn normal help circumv"
    [126] " problem"
    [127] ""
    [128] " machin learn problem direct map domain"
    [129] " learn use structur featur"
    [130] "chapter can learn normal"
    [131] "instanc observ distribut distanc centroid"
    [132] " observ specif token new request"
    [133] " can check token whether obtain distanc "
    [134] "previous calcul centroid statist sound rang"
    [135] "accord observ distribut sinc differ token web"
    [136] "request exhibit differ behavior centroid might fit"
    [137] " type token approach like markov model even simpl"
    [138] "lookup tabl can often appli success"
    [139] ""
    [140] " chapter present method attain level"
    [141] "respons flexibl waf name abil drop"
    [142] " also heal malici request without relianc known"
    [143] "pattern approach base anomali detect carri "
    [144] " granular http request token contrast previous"
    [145] "applic anomali detect web attack detect eg"
    [146] ""
    [147] " method detect react attack"
    [148] ""
    [149] " develop prototyp revers proxi call"
    [150] "implement idea mangl coupl anomali detect "
    [151] "prototyp http request pars tokenvalu pair "
    [152] "compar learn profil normal content specif"
    [153] "token token deviat typic profil "
    [154] "replac appropri benign valu use tokenspecif heurist"
    [155] ""
    [156] " propos request heal techniqu simpl effect henc"
    [157] " can effici implement deploy highspe"
    [158] "network main advantag socal request"
    [159] " simpl drop request decis made "
    [160] "precis context specif token instead full request"
    [161] "great improv detect accuraci verifi experiment"
    [162] "comparison detect request level make decis"
    [163] "faulttoler sinc replac content suitabl"
    [164] "altern certain case harm even wrong"
    [165] "classifi malici"
    [166] ""
    [167] " system high customiz especi reason"
    [168] "automat mangl may desir eg privaci reason one"
    [169] "may want automat replac user name password field"
    [170] "valu howev case manual configur "
    [171] "mandatori automat setup procedur provid "
    [172] "decis engin advantag abil"
    [173] "learn contamin enabl deploy"
    [174] "minim manual effort"
    [175] ""
    [176] " summari contribut chapter"
    [177] ""
    [178] " propos web applic firewal decid base anomalybas model part request anomal need replac benign part observ past"
    [179] " addit employ setup procedur assign type extract token statist featur"
    [180] " clean attackfre need sinc "
    [181] " learn model well test robust contamin"
    [182] " addit attack need learn"
    [183] " setup procedur"
    [184] ""
    [185] ""
    [186] " remain part chapter organ follow"
    [187] " introduc methodolog "
    [188] "selfheal web applic firewal present prototyp"
    [189] " evalu detect perform runtim use real"
    [190] "http traffic relat work discuss"
    [191] " conclus given"
    [192] ""
    [193] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file014.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[12]]
    $content
     [1] " token doctor"
     [2] ""
     [3] ""
     [4] " automat heal malici web request entail follow"
     [5] "two essenti task identif malici content "
     [6] "construct replac content unlik regular intrus"
     [7] "detect problem decid whether request"
     [8] "malici identif problem complex sinc"
     [9] "one need determin presenc malici content"
     [10] "also locat prerequisit replac"
     [11] ""
     [12] " identif problem can address make decis "
     [13] "refin context specif paramet http request"
     [14] "contextdepend detect previous use"
     [15] " uri paramet get request "
     [16] " full http request content follow"
     [17] "rough idea request pars tokenvalu"
     [18] "pair request discard point"
     [19] " instead combin score tokenbas model make"
     [20] "decis request level decis made token"
     [21] "level content token deem anomal"
     [22] "appropri replac sought"
     [23] ""
     [24] ""
     [25] ""
     [26] ""
     [27] " system act revers"
     [28] " proxi intercept client request examin"
     [29] " potenti alter request deliv product"
     [30] " system"
     [31] ""
     [32] ""
     [33] ""
     [34] " divers web applic traffic content essenti rule"
     [35] " possibl onesizefitsal traffic can"
     [36] "learn henc attempt provid right anomali detect"
     [37] "algorithm appropri heal action particular token"
     [38] " result design compris follow three orthogon"
     [39] "compon"
     [40] ""
     [41] ""
     [42] " type base analysi real http traffic"
     [43] " postul four token type describ characterist"
     [44] " distribut token content"
     [45] ""
     [46] " detector sinc malici content can manifest"
     [47] " various featur eg unusu length previous"
     [48] " unseen attribut differ anomali detect algorithm can"
     [49] " use captur attack employ anomali detector"
     [50] " automat coupl particular token setup"
     [51] " procedur present "
     [52] ""
     [53] " action particular transform token"
     [54] " content denot heal action propos four"
     [55] " heal action similar detector depend particular"
     [56] " token type also configur setup procedur"
     [57] ""
     [58] ""
     [59] " oper point view can character "
     [60] "revers proxi similar function traffic normal"
     [61] " intercept http client request "
     [62] "pars accord http protocol specif potenti"
     [63] "modifi request content relay traffic actual product"
     [64] "system contrast system propos "
     [65] " requir target product server run"
     [66] "differ secur level securityrel decis"
     [67] "encapsul proxi signific simplifi"
     [68] "design network environ protect"
     [69] ""
     [70] " architectur shown figur"
     [71] " exampl decis present "
     [72] "figur token uri paramet deem"
     [73] "normal remain unalt wherea anomal valu"
     [74] " flag heal exampl valu"
     [75] "automat replac previous seen benign highest"
     [76] "similar ie string correspond sql inject attack"
     [77] " replac benign string due occurr"
     [78] " string attack"
     [79] ""
     [80] " follow section provid detail descript "
     [81] "three orthogon compon follow present"
     [82] " setup procedur tie three compon togeth"
     [83] ""
     [84] " type"
     [85] ""
     [86] ""
     [87] "everi request receiv pars syntact part"
     [88] "accord http specif store tokenvalu"
     [89] "pair treat uri path paramet get post request"
     [90] " header field separ token follow respect"
     [91] "valu exampl request figur"
     [92] "pars five token correspond paramet header"
     [93] "denot insid revers proxi note request paramet"
     [94] " differ name will present differ token"
     [95] ""
     [96] " distribut token content high divers token"
     [97] "eg contain small number possibl valu"
     [98] "sometim even constant valu token queri"
     [99] " rang paramet contain wide varieti valu may"
    [100] "even generat automat order systemat handl"
    [101] " divers classifi token four type"
    [102] "accord typic properti inform transmit http"
    [103] "request"
    [104] ""
    [105] ""
    [106] " simplest case valu token"
    [107] " take valu exampl header field"
    [108] " monitor particular web host"
    [109] ""
    [110] " second type token carri "
    [111] " take small valu depend either http"
    [112] " protocol web applic exampl token"
    [113] " header"
    [114] ""
    [115] " input third type token compris"
    [116] " machinegener session number identifi"
    [117] " cooki"
    [118] ""
    [119] " input complex token type induc"
    [120] " human input freetext field queri string comment"
    [121] " name enter exhibit semant structur"
    [122] " except generat natur languag"
    [123] ""
    [124] ""
    [125] " characterist featur differ token type taken"
    [126] " account choic anomali detect algorithm heal"
    [127] "action"
    [128] ""
    [129] ""
    [130] " detector"
    [131] ""
    [132] ""
    [133] "anomali detect method wide studi protect"
    [134] "web servic"
    [135] ""
    [136] "howev previous approach flag anomali full http request"
    [137] " henc direct appli trigger finegrain"
    [138] "action individu token deploy pertoken anomali"
    [139] "detect algorithm propos howev"
    [140] "decisionmak remain token level"
    [141] ""
    [142] " choic anomali detect method depend token"
    [143] "type constant enumer token straightforward"
    [144] "occurr check refer list detector natur choic"
    [145] " given valu seen train deem"
    [146] "anomal remain two token type deploy three differ"
    [147] "detector describ decis detector "
    [148] "appli specif token automat made setup"
    [149] "process present "
    [150] ""
    [151] ""
    [152] " centroid anomali detector ncad"
    [153] ""
    [154] ""
    [155] "ngram model wide use secur applic"
    [156] " deploy"
    [157] "embed techniqu propos provid"
    [158] "effici way ngram analysi"
    [159] ""
    [160] "given possibl ngram byte sequenc w"
    [161] " 255 defin embed function w"
    [162] " token valu x "
    [163] ""
    [164] " w"
    [165] ""
    [166] ""
    [167] " return 1 ngram w"
    [168] "contain x 0 otherwis result vector"
    [169] "normal one elimin lengthdepend vector space"
    [170] "induc embed ngram grow exponenti n"
    [171] "howev spars linear length sequenc"
    [172] "allow one effici construct compar embed vector"
    [173] " byte sequenc detail "
    [174] ""
    [175] " embed function hand euclidean distanc"
    [176] "embed vector can defin follow"
    [177] ""
    [178] "dexz w"
    [179] ""
    [180] ""
    [181] "use distanc detect can perform comput"
    [182] "distanc previous learn normal"
    [183] ""
    [184] ""
    [185] ""
    [186] " de"
    [187] ""
    [188] ""
    [189] ""
    [190] " vector construct train x"
    [191] " xn arithmet mean respect embed"
    [192] "vector threshold"
    [193] " determin independ valid describ"
    [194] ""
    [195] ""
    [196] " chain anomali detector mcad"
    [197] " markov chain previous use sever"
    [198] "secur applic"
    [199] " use"
    [200] " 256 possibl byte valu state markov chain 256"
    [201] "possibl state transit see appendix"
    [202] "detail state transit probabl can learn record"
    [203] "transit frequenc byte bi bj train"
    [204] " includ extra start state overal size "
    [205] "transit tabl 2562 256 prohibit"
    [206] "larg learn transit probabl can estim"
    [207] " probabl token valu x length n base "
    [208] "learn markov chain"
    [209] ""
    [210] "p px1 x1"
    [211] " xi1xi xi"
    [212] ""
    [213] " xi correspond ith byte token valu x "
    [214] " use length normal instanc appli geometr"
    [215] "mean want detector take content length"
    [216] " account equip tokenspecif markov chain "
    [217] "threshold mcad new valu x defin follow"
    [218] ""
    [219] ""
    [220] ""
    [221] " p"
    [222] ""
    [223] ""
    [224] ""
    [225] ""
    [226] " anomali detector lad often length"
    [227] " token valu characterist attack exampl"
    [228] " major buffer overflow attack exhibit long token"
    [229] "valu properti address lad detector "
    [230] " solut token insuffici amount"
    [231] " render learn task involv train ncad mcad"
    [232] "imposs therefor find solut can cope "
    [233] "scarc situat modern robust statist provid us"
    [234] "power tool special deal noisi even"
    [235] "small sampl size especi small sampl size estim "
    [236] "mean standard deviat use chebyshev inequ"
    [237] "instanc can extrem outlier depend "
    [238] " statist base bias estim can loos"
    [239] ""
    [240] "henc instead use base chebyshev inequ"
    [241] "decid employ robust statist describ"
    [242] " given predefin signific level"
    [243] "estim 1 quantil length distribut"
    [244] " train valid l name l1"
    [245] " now construct confid interv l1"
    [246] " first calcul bootstrap estim "
    [247] "standard error l1 name"
    [248] "determin paramet c interv"
    [249] ""
    [250] " l1 c l1 c"
    [251] ""
    [252] " probabl coverag 1 final choos"
    [253] "upper bound confid interv threshold lad"
    [254] "detector allow futur variabl result "
    [255] "follow decis rule"
    [256] ""
    [257] ""
    [258] ""
    [259] " l1 c"
    [260] ""
    [261] ""
    [262] ""
    [263] ""
    [264] " action"
    [265] ""
    [266] ""
    [267] " finegrain detect token level allow us devis"
    [268] "similar finegrain heal action henc automat respons"
    [269] "mechan can less intrus accur action taken"
    [270] " request level particular equip "
    [271] "follow heal action"
    [272] ""
    [273] ""
    [274] " token conserv"
    [275] " respons spot anomal token valu remov"
    [276] " token request notic still much"
    [277] " benign action drop request use action"
    [278] " token lad detector"
    [279] ""
    [280] " encod altern still"
    [281] " conserv strategi encod anomal valu use html"
    [282] " entiti approach make common web attack base crosssit"
    [283] " script sql inject fail control punctuat"
    [284] " charact escap action provid almost damag"
    [285] " benign request mani web applic can resolv addit"
    [286] " encod content"
    [287] ""
    [288] " frequent valu"
    [289] " constant enumer token type natur heal action "
    [290] " replac valu frequent normal valu "
    [291] " token natur action assign token "
    [292] " list detector"
    [293] ""
    [294] " nearest valu "
    [295] " involv heal action replac anomal valu "
    [296] " nearestneighbor train replac possibl"
    [297] " due embed valu metric space introduc"
    [298] " default action"
    [299] " mcad ncad note sideeffect action can also"
    [300] " correct typo userinput field"
    [301] ""
    [302] ""
    [303] "clear four heal action tight coupl "
    [304] "particular type consid token precis assign"
    [305] " heal action token type present"
    [306] " also allow administr"
    [307] " tighten propos default action special token need"
    [308] "extra protect like password file cooki"
    [309] ""
    [310] " tokdoc"
    [311] ""
    [312] ""
    [313] ""
    [314] ""
    [315] " 12 0 10cliptruewidth75"
    [316] " test procedur setup "
    [317] " servicespecif split train test"
    [318] " procedur decid token detector "
    [319] " use exploit structur statist featur"
    [320] " automat process total datadriven x denot train specif token "
    [321] ""
    [322] ""
    [323] ""
    [324] "sinc main compon base learn method"
    [325] " setup depend avail initi corpus"
    [326] "normal train valid initi suffici"
    [327] "larg pool client request separ accord"
    [328] "servic eg virtual host andor differ web servic"
    [329] "allow servicespecif learn model pars"
    [330] "use generat tokenspecif pool use follow"
    [331] "phase amount chosen accord traffic"
    [332] "volum widest possibl rang normal behavior"
    [333] "cover note servicespecif split process "
    [334] " improv prisma framework introduc"
    [335] "chapter generat even focus"
    [336] "statespecif instanc"
    [337] ""
    [338] " test framework depict figur determin"
    [339] " token automat datadriven detector"
    [340] " assign exploit structur statist"
    [341] "featur use robust outlierresist statist procedur"
    [342] "ensur meaning decis even dirti attacktaint"
    [343] " set collect split two equal"
    [344] "size part train pool use learn "
    [345] " token threshold estim use valid"
    [346] " semiautomat assign action outlier"
    [347] "adjust threshold token system readi"
    [348] " deploy learn alreadi discuss"
    [349] " now describ detail"
    [350] " part setup process"
    [351] ""
    [352] " datadriven detector assign depict"
    [353] "figur step consist structur "
    [354] "statist carri token "
    [355] "origin start size procedur assign"
    [356] "simpl lad detector train current test"
    [357] "token contain 50 less sampl rational "
    [358] " detector need reason amount estim"
    [359] " model 50 sampl avail procedur"
    [360] "check whether current token enumer observ"
    [361] "less 10 uniqu valu token procedur test"
    [362] "statist evid exploit well known"
    [363] "first defin list c d train valid"
    [364] "describ whether sampl valid "
    [365] "observ train can defin function"
    [366] " return pvalu "
    [367] "whether c generat binomi variabl generat"
    [368] "fals probabl true probabl 1"
    [369] " now can determin maxim bare"
    [370] "support accept hypothesi c generat"
    [371] " given signific level"
    [372] ""
    [373] ""
    [374] ""
    [375] ""
    [376] " valu impress possibl nonmatch"
    [377] "occurr token might occur futur"
    [378] "similar can interpret upper bound confid"
    [379] "interv empir observ thus can use"
    [380] "valu threshold expect falseposit per token "
    [381] "list type"
    [382] ""
    [383] " decid ncad mcad procedur first look"
    [384] " structur featur name median length token sinc"
    [385] " ncad detector base 2gram detector need least two"
    [386] "charact calcul meaning mean distanc "
    [387] "token pass structur procedur focus "
    [388] "statist properti observ given centroid"
    [389] "largest distanc bound d"
    [390] " 1 sinc normal length"
    [391] "one use kernel densiti estim valid see"
    [392] " instanc detail can measur"
    [393] "bound probabl maxim distanc ever attain"
    [394] "formal px d 0"
    [395] ""
    [396] " ncad mcad need threshold oper sinc"
    [397] "model focus specif token can choos relat"
    [398] "relax threshold polici propos use maxim distanc"
    [399] " ncad minim probabl mcad semiautomat"
    [400] "outlier adjust valu valid "
    [401] " order accord output detector descend"
    [402] "distanc mean ncad ascend probabl mcad"
    [403] " administr decid whether extrem valu real"
    [404] "usergener sampl malici token valu exampl"
    [405] "sort probabl figur clear show"
    [406] " extrem three valu induc malici input"
    [407] "therefor administr adjust threshold first"
    [408] "usergener request"
    [409] " procedur administr addit can check"
    [410] "qualiti assign detector see whether chosen"
    [411] "fit actual "
    [412] ""
    [413] " addit can address privaci secur issu refin"
    [414] "assign action administr can manual adjust whether"
    [415] "token heal drop complet instanc privaci"
    [416] "relat cooki password must replac "
    [417] "nearest counterpart instead drop complet"
    [418] "prevent potenti abus like session password hijack"
    [419] ""
    [420] " system produc fals posit deploy can"
    [421] "track token caus fals alarm thus"
    [422] "administr can focus specif token can reconfigur"
    [423] "system accord incid case websit"
    [424] "restructur new servic deploy may "
    [425] " adjust accord potenti lead retrain "
    [426] "token model"
    [427] ""
    [428] ""
    [429] ""
    [430] ""
    [431] " setup consol automat detector assign administr check calcul threshold outlier"
    [432] ""
    [433] ""
    [434] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file015.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[13]]
    $content
     [1] ""
     [2] ""
     [3] ""
     [4] "evalu intrus prevent system multifacet"
     [5] "task sinc effect respons action inher depend"
     [6] " accuraci malici content identif first evalu"
     [7] " accuraci detector compar overal perform"
     [8] " stateoftheart method check realtim readi"
     [9] " runtim assess compar proxi"
     [10] ""
     [11] " perform"
     [12] ""
     [13] " evalu detect perform collect network"
     [14] "trace two differ internet domain first first08"
     [15] "compris 60 day traffic 1452122 http request record"
     [16] " web server research institut"
     [17] " 2008 server provid static content well dynam page"
     [18] "use content manag system second"
     [19] " blog09 cover 33day traffic 1181941 request"
     [20] " obtain domain run various weblog 2009"
     [21] "blog run popular publish platform"
     [22] " evalu set split three equal size"
     [23] "part train valid test due differ web"
     [24] "applic amount monitor token well assign"
     [25] " anomali detector differ set "
     [26] "configur use set present tabl"
     [27] ""
     [28] ""
     [29] ""
     [30] ""
     [31] ""
     [32] ""
     [33] " detector first08"
     [34] ""
     [35] " categori list lad"
     [36] " mcad ncad"
     [37] ""
     [38] " header 14 14 5 10 43"
     [39] " paramet 9 3 4 16"
     [40] " path 1 1"
     [41] ""
     [42] " 23 17 10 10 60"
     [43] " detector blog09"
     [44] ""
     [45] " categori list lad"
     [46] " mcad ncad"
     [47] ""
     [48] " header 22 77 15 17 131"
     [49] " paramet 14 166 28 7 215"
     [50] " path 1 1"
     [51] ""
     [52] " 36 243 44 24 347"
     [53] ""
     [54] ""
     [55] " configur use experi column categori summar token token origin header paramet queri path token introduc "
     [56] ""
     [57] ""
     [58] ""
     [59] " addit regular network traffic collect network"
     [60] "attack base 35 exploit obtain framework"
     [61] " well common secur archiv milw0rm packet"
     [62] "storm bugtraq attack execut virtual"
     [63] "environ thorough adapt characterist two"
     [64] " set detail list consid attack exploit"
     [65] " given tabl result variat"
     [66] " record eg usag differ shellcod encod sql"
     [67] "statement attack pool contain 89 attack instanc first08"
     [68] " 97 attack blog09"
     [69] ""
     [70] ""
     [71] ""
     [72] " l l"
     [73] ""
     [74] " cve milworm token detector"
     [75] ""
     [76] " buffer overflow attack"
     [77] ""
     [78] " mcad near 282930"
     [79] " mcad near 34353637"
     [80] " protocol viol parser struct drop 313233"
     [81] " protocol viol parser bodi drop 456"
     [82] " param normal drop 0123"
     [83] " protocol viol parser uri drop 2425"
     [84] " protocol viol parser struct drop 2627"
     [85] " protocol viol parser uri drop 71727374"
     [86] " param normal drop 636465"
     [87] " protocol viol parser uri drop 49505152"
     [88] " mcad near 4546"
     [89] " mcad near 686970"
     [90] " list freq 20212223"
     [91] " ncad near 2023"
     [92] ""
     [93] " code inject attack"
     [94] ""
     [95] " param normal drop 10111213"
     [96] " param normal drop 56575859"
     [97] " mcad drop 606162"
     [98] " param normal drop 171819"
     [99] " mcad drop 4748"
    [100] " mcad near 789"
    [101] " php param normal drop 8285"
    [102] " sql param s mcad enc"
    [103] ""
    [104] " wordpress attack"
    [105] ""
    [106] " mcad drop 101"
    [107] " param mcad near 102 103"
    [108] " mcad drop 105"
    [109] " param mcad near 104"
    [110] " param normal drop 108"
    [111] " param normal drop 110"
    [112] " param lad drop 109"
    [113] " param normal drop 107"
    [114] " param lad drop 106"
    [115] " ncad near 100"
    [116] ""
    [117] " miscellan attack"
    [118] ""
    [119] " httptunnel protocol viol parser struct drop 868788"
    [120] " protocol viol parser uri drop 6667"
    [121] " protocol viol parser struct drop 535455"
    [122] " xsssql param mcad enc 75767778"
    [123] ""
    [124] ""
    [125] " http attack"
    [126] "tabl http exploit attack execut differ variant"
    [127] " list token attack locat detector case token never seen normal request drop token"
    [128] ""
    [129] ""
    [130] ""
    [131] " learner request semant"
    [132] ""
    [133] "first want check whether differ detector model"
    [134] "realli necessari construct special instanc"
    [135] " just lad detector instead mcad ncad"
    [136] "refer just ncad ie mcad"
    [137] " replac ncad refer mcad"
    [138] "ie ncad replac mcad refer "
    [139] " evalu instanc "
    [140] " first08 blog09 reject request manual"
    [141] "check label fals true posit case doubt"
    [142] "request replay origin server follow first"
    [143] "use unmodifi request save repli server "
    [144] "compar repli server send request modifi"
    [145] " differ count request fals"
    [146] "posit complet replay process observ"
    [147] " sever drastic repli server indic malform"
    [148] "even malici request prove inher request"
    [149] "semant harm action "
    [150] ""
    [151] ""
    [152] ""
    [153] ""
    [154] " detector fp tp fn"
    [155] ""
    [156] " first08 000002 0 000000"
    [157] " 000000 0 002247"
    [158] " 000001 0 000000"
    [159] " 000002 0 022472"
    [160] ""
    [161] " blog09 000003 212 004124"
    [162] " 000001 68 015464"
    [163] " 000009 186 004124"
    [164] " 000003 0 022680"
    [165] ""
    [166] ""
    [167] " perform sever instanc fp falseposit rate tp attack found normal traffic fn falseneg rate"
    [168] ""
    [169] ""
    [170] " result summar tabl first"
    [171] "thing notic overal low falseposit rate "
    [172] "direct result addit pars local decis make"
    [173] " closer look reveal "
    [174] " suffer high falseneg rate "
    [175] " perform equal good first08 "
    [176] "fall behind involv blog09 plain"
    [177] " divers model method"
    [178] "perform near ident set also capabl"
    [179] "detect true posit taint blog09 "
    [180] "trend confirm tabl"
    [181] "present detector necessari disarm use attack note"
    [182] " malici part attack spread throughout differ"
    [183] "token render approach even"
    [184] "valuabl addit heal action employ "
    [185] "save rough 00001 discard fals"
    [186] "posit blog09 first08 summari"
    [187] "result show just full varieti model embodi"
    [188] " lead overal good perform keep"
    [189] "general request semant intact"
    [190] ""
    [191] " detector"
    [192] " baselin detect perform consid two"
    [193] "stateoftheart anomali detect techniqu use raw http request"
    [194] "payload input chain detector use markov"
    [195] "chain describ full content"
    [196] " request anomali detect learn train"
    [197] " similar calibr use valid"
    [198] "partit second baselin implement variant"
    [199] " detector store ngram"
    [200] "benign http request bloom filter use ratio unknown"
    [201] "ngram incom request anomali score detector"
    [202] "calibr valid n fix 2"
    [203] ""
    [204] " result evalu summar"
    [205] "tabl first08 blog09 report"
    [206] " equal falseposit rate detector"
    [207] "calibr trueposit rate "
    [208] " rate miss regular attack detector"
    [209] " calibr falseposit rate "
    [210] ""
    [211] ""
    [212] ""
    [213] ""
    [214] " detector"
    [215] ""
    [216] " first08 000002 000000"
    [217] " markov chain 002005 080899"
    [218] " anagram 000004 016854"
    [219] ""
    [220] " blog09 000003 004124"
    [221] " markov chain 016698 018557"
    [222] " anagram 100000 039175"
    [223] ""
    [224] ""
    [225] " perform payloadbas anomali detector falseposit rate detector calibr trueposit rate rate miss regular attack detector calibr falseposit rate "
    [226] ""
    [227] ""
    [228] "focus first08 see anagram"
    [229] "yield accept falseposit rate howev anagram much"
    [230] "porous near 17 attack detect contrari"
    [231] " capabl detect attack attain even"
    [232] "lower falseposit rate anagram markov chain"
    [233] "simpli overburden first08 even blog09"
    [234] " falseposit rate rise unaccept"
    [235] "17 surpris anagram break blog09 "
    [236] " calibr trueposit anagram flag"
    [237] " legitim request anomal due fact"
    [238] " 23 attack anomali score 0 "
    [239] "smallest possibl score attain therefor tag incom"
    [240] "request anomal even calibr anagram 23"
    [241] "falseneg rate rough 8 time higher "
    [242] "setup yield still falseposit rate 000038 "
    [243] "magnitud higher number clear demonstr"
    [244] "outstand perform term fals posit"
    [245] " negat even hard set like blog09"
    [246] ""
    [247] " perform"
    [248] ""
    [249] " deliv inlin intrus prevent system "
    [250] "reason fast sinc everi client request pass revers"
    [251] "proxi without intoler delay part subject"
    [252] " prototyp stress see whether can use "
    [253] "realtim scenario"
    [254] ""
    [255] " prototyp implement use "
    [256] "framework framework provid matur interfac number"
    [257] "network protocol reus proxi modul integr"
    [258] "optim ngram c librari abl produc"
    [259] "fullfledg prototyp system replay complet"
    [260] "test slice first08 blog09 approxim 500k"
    [261] "request get stabl estim process time "
    [262] "baselin measur process time "
    [263] "proxi second consid web applic firewal"
    [264] " minim setup rule assess addit pars"
    [265] "affect process time request furthermor use "
    [266] "simpl forward proxi applic implement "
    [267] "framework see much framework impos "
    [268] "process runtim final "
    [269] "environ median runtim setup present"
    [270] "tabl"
    [271] ""
    [272] ""
    [273] ""
    [274] ""
    [275] ""
    [276] " proxi"
    [277] ""
    [278] " modsec"
    [279] ""
    [280] " first08 1387 1536 2552 2768"
    [281] " blog09 1500 1694 2430 2902"
    [282] ""
    [283] ""
    [284] " runtim millisecond differ proxi"
    [285] " first08 blog09"
    [286] ""
    [287] ""
    [288] ""
    [289] "first can observ two set exhibit differ"
    [290] "baselin general first08 seem simpler"
    [291] "structur compar blog09 furthermor high"
    [292] "optim rough 1 ms per request faster"
    [293] "compar equival look "
    [294] "interappl differ can observ increas 01 ms"
    [295] " 02 ms 02 ms 05 ms "
    [296] " proxi respect impli "
    [297] "addit anomali detect method employ just add"
    [298] " rough 01 ms 03 ms per request experi clear"
    [299] "demonstr still room improv term"
    [300] " runtim anomali detect method use suitabl"
    [301] " run inlin system even current"
    [302] "unoptim state alreadi can use intrus prevent"
    [303] "system"
    [304] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file016.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[14]]
    $content
     [1] " work"
     [2] ""
     [3] ""
     [4] " automat protect web applic gain increas"
     [5] "attent among secur research convent id "
     [6] " reli specif attack"
     [7] "signatur predefin attack characterist provid"
     [8] "adequ time protect dynam chang web attack"
     [9] "anomali detect techniqu base payload analysi exampl"
    [10] ""
    [11] "provid possibl detect previous unknown attack"
    [12] " approach enabl protect differ network servic"
    [13] "attain suffici throughput rate yet lack protocol context"
    [14] " analysi restrict use intrus prevent"
    [15] "simpl drop redirect packet"
    [16] ""
    [17] "first protocolawar method detect attack web traffic"
    [18] "use anomali detect propos "
    [19] "extend ensu work"
    [20] " main idea"
    [21] " method combin multipl anomali detector"
    [22] " length check byte distribut hidden markov model appli"
    [23] " individu uri paramet similar finit state automata"
    [24] " multipl markov chain"
    [25] " recent propos detect anomal http request"
    [26] " approach differ method detect"
    [27] "anomal content individu token instead combin"
    [28] "tokenlevel anomali estim judg anomali complet"
    [29] "request finegrain detect enabl us devis novel token"
    [30] "heal action much less disrupt request drop"
    [31] ""
    [32] "anoth line research combin network anomali detect host"
    [33] "monitor propos system "
    [34] "anomal request execut special instrument shadow"
    [35] "honeypot system feedback whether request actual harm"
    [36] "system can use updat anomali detector similar"
    [37] " work line idea"
    [38] " combin sql attack detect revers proxi"
    [39] "forward request web server manag differ level"
    [40] "sensit inform depend anomali valu web"
    [41] "request sql queri anomali detector host decid whether"
    [42] " model web request anomali detector updat"
    [43] " request result malici databas queri "
    [44] "requir addit host instrument serv "
    [45] "transpar proxi great simplifi practic deploy"
    [46] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file017.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    $content[[15]]
    $content
     [1] " conclus"
     [2] ""
     [3] ""
     [4] " introduc protocolawar revers proxi "
     [5] "capabl decid token level part request"
     [6] " deem normal anomal sever intellig mangl"
     [7] "strategi anomal token apart just drop "
     [8] " describ experi realworld set demonstr"
     [9] "use approach runtim measur show readi"
    [10] " inlin intrus prevent"
    [11] ""
    [12] " prototyp show good perform especi term"
    [13] "fals negat awar sever extens can improv"
    [14] " extend system practic consider includ"
    [15] "integr platform"
    [16] " valuabl step toward runtim improv coupl"
    [17] " system shadow system propos"
    [18] " incorpor feedback loop "
    [19] "similar vein combin learn"
    [20] "techniqu anoth promis extens case "
    [21] " abl leverag knowledg under protocol "
    [22] "effici preprocess pars incom request"
    [23] "accord http grammar complex web applic"
    [24] "web server serv multitud servic might "
    [25] "signific overlap differ use token type accord "
    [26] "under state servic differ type servic"
    [27] "host thus applic method chapter"
    [28] " layer respons mechan benefici"
    [29] " session track state servic accord "
    [30] "infer abstract state machin anomali model sequenc "
    [31] "depend state even finer distinct abnorm"
    [32] "behavior achiev mechan also amount "
    [33] "integr sessionawar long term memori "
    [34] " use flag user session lot"
    [35] "anomal token seen danger use"
    [36] "reliabl close suspici connect even track attack"
    [37] "distribut sever request"
    [38] ""
    [39] "overal system proven promis fullfledg"
    [40] "web applic firewal present state capabl"
    [41] "effect prevent heal wide rang recent"
    [42] "webbas attack runtim perform make readili applic"
    [43] " protect modern web applic see layer"
    [44] "structur process toolchain pars request"
    [45] " emb token specif chosen local represent"
    [46] " enabl subsequ check statist signific"
    [47] "aberr learn normal model"
    [48] "approach even taken step sens map back"
    [49] " outcom probabilist meaning"
    [50] "healingreact act transform inform"
    [51] " probabilist domain back real world layer"
    [52] "structur approach permit stepwis develop"
    [53] "success applic probabilist method realworld"
    [54] "problem"
    [55] ""
    [56] ""
    [57] ""
    [58] ""
    [59] ""
    [60] ""
    [61] " chapter summar find thesi start"
    [62] "review main part analysi detect respons"
    [63] "framework develop preced chapter inspect"
    [64] " individu compon framework relat "
    [65] "frequentist view statist gain insight "
    [66] "precondit applic postcondit probabilist"
    [67] "method network secur conclud chapter outlook"
    [68] " implic find relat classic"
    [69] "insight softwar engin potenti pitfal"
    [70] "merit realworld applic"
    [71] ""
    
    $meta
    $author
    character(0)
    
    $datetimestamp
    [1] "2014-06-11 21:28:01 GMT"
    
    $description
    character(0)
    
    $heading
    character(0)
    
    $id
    [1] "file018.tex"
    
    $language
    [1] "en"
    
    $origin
    character(0)
    
    attr(,"class")
    [1] "TextDocumentMeta"
    
    attr(,"class")
    [1] "PlainTextDocument" "TextDocument"
    
    
    $meta
    list()
    attr(,"class")
    [1] "CorpusMeta"
    
    $dmeta
    data frame with 0 columns and 15 rows
    
    attr(,"class")
    [1] "VCorpus" "Corpus"
    > thesis = corpusToPrisma(thesis, NULL, TRUE)
    Loading required package: tm
    Error: inherits(doc, "TextDocument") is not TRUE
    Execution halted
Flavor: r-oldrel-windows-ix86+x86_64