http://kingman.cs.ualberta.ca/_banff04/icml/pages/accepted.htm
ICML 2004
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/321.pdf	63	Multiple Kernel Learning, Conic Duality, and the SMO Algorithm	acknowledge acknowledgements adaptive addition advances algorithm algorithms also andersen applications approximately aspects athena bach bartlett because bertsekas bhattacharyya bousquet boyd bradley brent bret calculation cambridge canu chapelle check choosing classi complexity computation concave conditions cone conference continuously convex convexity corollary corporation cristianini denoted derivative derivatives design done each easily easy endix englewood enough entiable equal erentiability erentiable estimate fast feature fellowship francis from function ghaoui given graduate grandvalet grant hall high homogeneous hyperkernels implementation implies improvements information inspecting intel interior international joachims jordan kaufmann keerthi kernel lanckriet large learning lemarechal linear lobo machine machines making mangasarian mateo matrix maximize methods microsoft minimal minimization minimize minimum moreau morgan mosek mukherjee multiple murthy nashua need neural nite nonlinear optim optimal optimality optimization optimizer order otherwise over pair parameters partial perf piecewise platt point points practical preliminaries prentice press processing programming prove provide quadratic references regularization regularized requires research respect reveals sagastizabal scale scaling scienti second selection semide sequence sequential shevade short shows siam small smola sorting strict strictly such support svms systems that then theoretical this thus training unique uniqueness used using vandenberghe vapnik vector weights where which williamson wish with without yosida zero
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/348.pdf	38	Gaussian Pro cess Classification for Segmenting and Annotating Sequences	advances algorithm algorithmic algorithms altun approximation based bennett between boosting classi collins computation conference crammer csat data discovery discriminative embrechts emnlp empirical equivalence experiments gaussian gibbs girosi heterogeneous hidden hofmann ieee implementation information international johnson journal kernel know label language learning ledge line machine machines mackay mark markov methods mining models momma multiclass natural neural opper perceptron proceedings process processes processing references report research sequences sigkdd singer sparse support systems technical theory training tsochantaridis variational vector with
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/45.pdf	0	A Comparative Study on Metho ds for Reducing Myopia of Hill-Climbing Search in Multirelational Learning	algorithmic applications applis arti blockeel bratko cation cial clauses cohen computation conf constraint data decision detecting discretization down element eroski gleton handling induction inductive intel introduction jacobs jezernik laer lavra learning ligence logic logical lookahead machine methods mining molina moure mugz nite noise order problems proc programming prolog raedt recursive references relational theory trees with workshop
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/259.pdf	58	Linearized Cluster Assignment via Sp ectral Ordering	advances aggregation aided algorithm amer analysis bach barnard cadintegrated chan chung circuits cluster clustering component computed computer conf cuts data desgin ding discovery envelope european graph hagen hierarchical ieee image info information intel jordan kahng know learning ledge ligence machine malik math matrices means meila merge methods mining multiclass multiway neural nips normalized partitioning pattern pdkk pothen principal principles proc processing ratio reduction references relaxation report scaled schlag segmentation self simon society space sparse spectral split supercomputing systems tech theory trans unsupervised vision washington weiss zien
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/119.pdf	42	Integrating Constraints and Metric Learning in Semi-Supervised Clustering	ablating accurate acknowledgments adaptive advances advantages algorithm algorithms allows along also annie anonymous application approach approximation arti available avenue background banerjee bansal based basic basu bennett bilenko bilmes bioinformatics blake blum both california cardie caruana cation centroids chawla cial classi cluster clustering clusters cohn combined comments compared comparisons components computer conclusions conference constrained constraining constraint constraints contrast cornell correlation costs data databases datasets demiriz detection diagonal different dimensional discovering discovery dissertation distance doctoral domeniconi duplicate elds embrechts engineering equivalence estimation euclidean experimentally expression extending facilitate feature feedback focs foundations framework from full functions future gaussian gene generation genetic gentle globally good grants handling have hertz hidden high hillel html http icml icsi ieee improvements include individual individually information informative initial initialization insightful instancelevel integrated interaction interesting international isolation joachims joel jordan journal kamvar klein kleinberg knowledge koller labeling lead learnable learning level like locally machine making manifolds manning markov matrix mccallum means measures merz methodology methods metric metrics mining mixture mlearn mlrepository models molecular mooney most networks neural nineteenth ninth noisy obtaining other outperforms over pairwise paper parameter part pathways pattern performing performs poorly presented previous previously primary prior probabilistic problems proceedings processing proposed protein publication random references relations relationships relative report repository research results reviewers riverside rogers roweis russell saul schroedl schultz science seeding segal selection semi semisupervised sensitive shapes shental shown sideinformation sigkdd similarity space standard string submission subspaces supervised supported symposium systems tardos tech techniques thank that think this topics tropp tutorial unifying university unlabeled unsupervised user uses using utexas variation wagstaff wang weight weighting weinshall well where with work would xing
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/229.pdf	27	Delegating Classifiers	accurate advances algorithm among analysis arbitrating area argamon arti based blake boosting brazdil cascade cation cial clark class classi competing conf conquer curve data databases domingos environments estimation european evaluation experiments fawcett ferri flach frank freund from furnkranz gama generalisation generalization generating global grading hand hernandez icml implementations imprecise improving induction information intelligence intelligent java kaufmann knowledge koppel learn learned learning logical mach machine merz mining morgan multiple networks neural niblett nitions optimization ortega practical probabilistic probability problems proc provost publishers quinlan ranking referees references relations repository review robust rule schapire seewald separate sets simple stacked systems techniques till tools tree trees under using with without witten wolpert
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/2.pdf	12	Active Learning of Lab el Ranking Functions	advances algorithm angluin approach arti available bertsimas bodies boundaries cambridge campbell cation cavtat chang chapter cial cjlin classi cohen committee computing concept conference constraint convex crammer cristianini croatia csie decision distributions dubois european foundation freund furnkranz graepel grunbaum herbrich http hullermeier hyperplanes icml information intelligence international journal large learning library libsvm machine machines margin mass math montreal multiclass neural nips obermayer operational order ordinal paci pairwise partitions peled possibility prade pranking preference press proceedings processing programs qualitative queries query random rank ranking references regression research roth sabbadin sampling schapire selective seung seventeenth shamir singer smola software solving springer stoc support symposium systems theoretic theory things tishby using vector vempala verlag walks with zimak
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/189.pdf	6	A Monte Carlo Analysis of Ensemble Classification	adaboost algorithmics analyzing annual arti bagging boosting brassard bratley bregman breiman carlo cial collins computational conference dipartimento dissertation distances doctoral eighteenth ensemble esposito explanation framework freund game hall informatica intel international italy joint kaufman learing learning ligence logistic machine monte morgan online practice prediction predictors prentice proceeding proceedings publishers references regression saitta schapire singer svizzera theory torino universit
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/85.pdf	51	Learning a Kernel Matrix for Nonlinear Dimensionality Reduction	academy advances arising arts based belkin bengio biswas burges cambridge clustering completion computation cortes data department dimensional dimensionality distributed donoho eigenmaps electrical embedding engineering extensions from geometry graepel grimes hessian high information inter invariance isomap kernel kernels laplacian learning linear localization locally matrix method methods mohri national network neural nite niyogi paiement paper press proceedings processing programming programs rational reduction references representation sample sciences semide semideinite sensor solving spectral stanford support systems techniques university vector vincent wireless working
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/100.pdf	62	Margin Based Feature Selection - Theory and Algorithms	according acting adaptive algorithm algorithms also analysis annals annual anthony application apply approach approximates arti ascent assigns aviation bachrach ball bartlett based been begin bellman benavente between bialek boosting bottleneck bound boundary bounded calculate calculated cation challenge chapelle choices choosing cial cient class classi communication complementary component compute computer computing conference consistency control cortes cover crammer criterion data database decision dependent derived dietterich dimension dimensionality directions discrimination discriminatory distance distribution does ectiveness elissee embedding every explanation face fath feature features following freund from function functions further generalization generated gilad globerson gradient guided gunn guyon hart have hence here hierarchies hodges http ieee important induction information intel international introduction john jolli journal just kaufmann kira known kohavi kohonen label labeled largest learnig learning lemma lerton ligence limitations line linear lipschitz local locally machine magnitude main maps margin martinez maximizes maximizing measure measuring medicine method methods minimization more morgan mukherjee naive navot nearest need neighbor network networks neural nips nipsfsc nonlinear nonparametric norm obtain omitted optimization order organizing originally over packing pattern pereira perform performs poggio points pontil practical presented press princeton principal probability proc processes processing projection proof properties prove proving published quality quinlan radius readings real reduction references relief rendell report representatives research result risk roweis rule rules sample saul schapire school science sciences search searching selection self shattering shavlik shawe showed shows sign simba simple since size soton space springer stated statistics straightforward structural subset such suggested summary support supported svms system systems taylor tech technical techniques than that then theorem theoretic theory therefore these this thus tishby tool tour trans transactions trees turn union university usaf using valued vapnik variable variety varlag vector verlag voting well weston which while wieghts williamson with workshop wrapper
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/69.pdf	47	Learning Bayesian Network Classifiers by Maximizing Conditional Likeliho o d	advances agresti algorithm allen analysis approach approaches approximating approximations arti augmented bacchus baltimore based bayes bayesian belief bias bilmes blake bound brandman byrne california cambridge categorical cation causation center chickering chow cial cient classi combination comparison computational computer conditional conf cooper criteria curse data databases dependence dept dimensionality discovery discrete discrimination discriminative discriminatively distribution distributions domingos duda empirical entropy environments extension fawcett feature filali flannery francisco friedman from geiger generative glymour goldszmidt graphical greiner hart hastie heckerman herskovits hidden holtz hopkins html http ieee imprecise induction inference information informative intel intl irvine jaakkola jackson jebara john johns jordan kaufmann keogh know knowledge kohavi language learning ledge ligence ligent likelihood livescu logistic loss machine marginal maximization maximum meila merz method mining mlearn mlrepository model models morgan naive natl nets network networks neural numerical optimality parameter pattern pazzani pearl pentland plausible prediction press principle probabilistic probability proc processing provost ranking reasoning recipes recognition references regression repository rept richardson robust rubinstein sandness scene scheines science search selection simple speech spirtes springer statistical statistics structural structured subset systems tech teukolsky theory torres transactions tree trees tutorial under univ university vapnik variables variance vetterling wiley with wkshp wrappers york zero zhou zweig
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/82.pdf	57	Leveraging the Margin More Carefully	adaboost adaptive advances advantage alas algorithm algorithms american appears application articles association bagging bartlett base baxter become behavior better between binary boost boosters boosting bounds bregman cases cation checked choose choosing classi close collected collection collins colt combining compared comparison complexities computation computer consists constructing contributed convex criterion data dataset decision dekel dempster dence dietterich difference different digits discriminate discussed dissertation distances divided doctoral document documents duffy each ensembles entire entropy entry epsilon error eurocolt experimental experiments explanation features figure four frean freund from function functional gaussian generalization gets given global gradient grows handbook helmbold hoffgen horn horst hypotheses icml ieee improved improves incomplete indistinguishable information insensitive international into iterations jcss labeled lack laird lang large latent learning least left less likelihood line logistic loss lower lter machine majority malicious many margin mason maximum mccallum mendelson methods mitchell much netnews neurons newsgroup newsgroups newsweeder nigam nips noise number obtained often only opportunity optimization otherwise outliers outperforms over pairs parallel parameters pardalos partially plan possible potential potsdam predictions principle prone psilearning rademacher randomization randomly rate rated ratsch references regression representing result results right risk robust rosenfeld rounds royal rubin schapire schuurmans sciences sequential servedio shalev shen show shwartz side simon singer single small smooth society some space statistical structural symmetrization symposium system tasks techniques test text than that then theoretic theory there this thousand three thrun thus topics trainability training trees tseng type university unlabeled usenet using usps vanilla various vector version wang weak were when which with wong word words zhang zhao
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/231.pdf	69	Redundant Feature Elimination for Multi-Class Problems	algorithm almuallim analysis appice application approach artificial attributes based bayes bhattacharyya blum boosting ceci class classification classifier cohen computation computer conf correlation dash data databases decision design dietterich discovery discrete dzeroski effective estimating examples extensions fast feature features freund generalization hall improvements induction intelligence intelligent irrelevant john kaufmann kira knowledge kohavi kononenko langley lavra learning line machine malerba many mining morgan multi murthy naive neural numeric pfleger platt practical practice press principles problem proc references relational relevant relief rendell rule schapire sciences selection shevade springer subset system theoretic verlag with
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/132.pdf	67	Optimising Area Under the ROC Curve Using Gradient Descent	aaai above accurate advances affect after against algorithms almost although analysis appropriate approximation area automatic avoided bamber based been below between binarisation binary boosted bradley cambridge cantly cases categorization centroid characteristic claim class classi cliffs compared comparison conclusion conf conference considered continuous cortes creates cristianini curve data decision descent development different digit direct distribution document dodier dominance ecml effect either empirical enabling englewood enough error european evaluation experiments explore extended extending extension fast favour features feedback figure focussed found from future gradient gradients graph hall have include increasing induction information interest intl introduced introduction issue joachims journal kernel laerning large larger learning linear logistic machine machines made mann many math measured menlo method methods minimal minimization models mohri more mozer neural number observations only operating optimise optimises optimization optimizing ordinal other outperform over park pattern performance perlich plan planned platt polynomial prediction predictors prentice press proceedings processing provost psych quite random rank rankopt rate receiver recognition references regression rejected relevance relevant report research retrieval rocchio rutgers scaling section sequential series shawe shown sigmoid signi simonoff small smart space statist statistic statistical steep stochastically strictly study stumps support surface system systems tasks taylor technical techniques tenth test text than that their them theory therefore these this thoroughly thousands trade training tree true twentieth under university using valued values vapnik variables vector weather weiss whereas whether which whitney wilcoxon wiley will with witney wolniewicz work would york
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/98.pdf	24	Communication complexity as a lower b ound for learning in games	acapulco adaptive against agent agents algorithm algorithms annual approach approximation arti awesome bargaining bayes behavior blum bowling brown cambridge canada cial cing commerce communication complexity computing conference conitzer continuation contributions convergence converges correlated dels dilemma distributed dominated drich dynamics eated econ eighteenth electronic eliminating endent equilibria equilibrium erate erative framework freund fudenb game games general gilb gradient graphical greenwald hall hannan ijcai indep information intel international internet joint kalai kearns koller kushilevitz learning learns levine leyton ligence littman machine mansour markov mathematics metho mexico multi multiagent multiplicative nash neural nips nisan olynomial onents onse operation optimal papadimitriou play playing press proceedings processing questions rate references reinforcement related research resp risk sandholm satis schapire self shelton singh some stationary stimpson stoc stone strategies structured symposium systems team tennenholtz that theoretical theory time uncertainty university using vancouver variable veloso wang weights wellman zemel
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/178.pdf	54	Learning to Learn with the Informative Vector Machine	advances algorithm also applied apply approach approximations arises arti aston available average baxter bayesian being belong beyond biol blake block cambridge caruana case cation cial ciency cient ciently circles classi colt combined computational compute conclusions considering covariance crosses csat dashed data databases dependent detect diagonal diekhaus discriminative dissertation doctoral does dordrecht dotted each easier entropy error estimate every expressing fast figure fisher from full gaussian generalised generative given graphical hasselmo haussler have herbrich heuristic homologies http independent information informative intel internal iterative jaakkola kaufmann kernel kernels kluwer large lawrence learn learning likelihood line linear machine makes matrix maximum merz method methods minimise minka model modelled models more morgan mozer much multi multiple multitask netherlands neural optimised order parameters phoneme picard pluses point points posterior prediction press proc process processes processing protein publishers random rasmussen rate rather recognisers recognition references regression remote repository representations revised same sampled sampling seeger selects sets should showed shown solid sparse speaker standard standpoint stat statistical suggesting systems task tasks tests than that then thing this thrun time touretzky train training underlying university used using vector vocabulary where whose williams with word
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/134.pdf	19	Bayesian Inference for Transductive Learning of Kernel Matrix Using the Tanner-Wong Data Augmentation Algorithm	advances algorithm algorithms alignment amari baltimore boosting bousquet cambridge complexity computations crammer cristianini data dempster design edition elissee from geometry golub herrmann hopkins incomplete information johns journal kandola kernel keshet laird learning likelihood loan matrix maximum networks neural press processing references royal rubin series shawetaylor singer society statistical systems target third university using
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/335.pdf	14	Apprenticeship Learning via Inverse Reinforcement Learning	abbeel after algorithm algorithms also alvinn amit application applying apprenticeship assuming atkeson autonomous bound bounded cauchy cauchyschwarz cient class coincide coincides combination completes component considered construct controller convex coordinate current decisions demiris demonstration denote dimension dimensional ding distance each easily estimate expectations extracting factor feature figure from full given gives guarantees harada have hayes hogan http human icdl icml imitation implies inaba inequality inoue invariance inverse iterate iteration iterations jection jectories keeping keeps knowledge kuniyoshi land learning least lemma line linear long management manne margin mataric maximum mind movement movements network neural neuroscience nips nition nitions norms notation note number observation obtained onto order organizing origin pabbeel paper performance point points policy pomerleau previously principle probability probaility proc programming progress projection proof prove proves recall reduced references reinforcement return reusable reward rewrit rewriting rewritten robot russell same sample satis schaal schwarz science sequences sequential sets shaping simplicity since single smaller spaces stanford step substituting such system take task terms that their then theorem theory this through thus tion track transformations under union used using vectors vehicle veri visual voluntary watching where which will with
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/94.pdf	11	Active Learning Using Pre-clustering	active advances aggregated algorithm analysis anniversary applications arti association based bayesian both boundary campbell categorization cation chang chapelle chen cial classi closest cluster clustering cohn computational conf conference content cristianini data database databases detection development digits documents ecir edinburgh employing error estimation european experiments experts face feedback figure finding framework from gale ghahramani groups handwritten hard hastie hubert ieee image images import information integrating intelligence international introduction john jordan journal kaufman kaufmann kernel kernels koller labeled labelled language large learning less letters lewis linear linguistics logistic machine machines margin mccallum meeting methods miller mitchell mixture mnist models more morgan multimedia natural network neural nigam number oles optimal ottawa parsing pattern percentage pham philadelphia plus pool probability problems proc proceedings processing proposed query random recogn reduction references regression regularized report representative research respectively results retrieval robust roukos rousseeuw sampling schohn scholkopf seeger semi sequential shen sigir size smeulders smola sons springer statistical statistics struyf supervised support systems tang technical techniques text through thrun tong toward training trans trec tresp uiuc university unlabeled unlabelled using uyar value vector verlag wang weston wiley with worring zhai zhang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/297.pdf	68	Predictive Automatic Relevance Determination by Exp ectation Propagation	accuracy achieve adaboost advances against algorithm algorithms also among analysis applications approximate ardep argues automatic balance bayes bayesian been benchmark best better between bioinformatics breast campbell cancer cation centre choose chooses choosing cient classi classifying comparable comparably computation conclusions continuum critical data dataset datasets desired determination diabetes distinct domain elissee error estimate estimating evidence expectation expression faul feature features first fraction framework from function gaussian gene generalization graepel guyon herbrich highdimensional html http hyperparameters ijcai inference information instead institute interpolation issue jmlr journal kernel large larger largest lead learning leave lecture likelihood ljubljana loss lower machine mackay marginal margins maximized maximizing mean medical methods micorarray minimizes minka model models most muller ndings neal networks neural neurocolt nips nonzero note notes number obtained oncology only onoda opper optimized other outperforms over papers performance performs point points popular possible practice predictive press principles processes processing propagation references relevance report research resulting results second selection show similar small soft soklic space sparse sparser sparsity special springer stat state stateof statistics success svms systems technical test testing than that theory this time tipping trading tsch tuning uncertainty university used useful using variable vector where which winther with workshop york yugoslavia zwitter
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/308.pdf	30	Dynamic Conditional Random Fields: Factorized Probabilistic Models for Labeling and Segmenting Sequence Data	aaai abbeel about abstract accurate acoustics advances algorithms american analysis annotated annual applications approach approaches approximate arti association audio automation based bayesian belief berger berkeley brill buchholz building cambridge causation chapter chunking cial comparison computational computer conditional conf conference conll convergence corpora corpus coupled craven croft cycle cycles data dean decision decoding discriminative dissertation doctoral dynamic eighteenth elds emnlp empirical english entropy erikt estimation extraction factorial family fifteenth fine foundations francisco freeman freitag frietag from generalized geometric ghahramani graphs hidden hierarchical hmms horn html http human ieee ijcai inference information intelligence international introduction iterative jaakkola joint jordan journal kanazawa kaufmann koller kudo labeling lafferty language large learning liang linear linguistics loopy machine machines mahadevan malouf manning marcinkiewicz marcus markov matsumoto maximum mccallum mceliece methods minka model models mohri morgan murphy naacl national natural navigation network networks neural nips nite north observable ofspeech papers parameter parsing part partially paskin peng penn pereira persistence peshkin pfeffer pietra pinto policy press probabilistic proc proceeding proceedings processes processing propagation rabiner ramshaw random ratnaparkhi reasoning recognition references relational reparameterization representation research riley robot robotics rohanimanesh rule sang santorini schutze segmentation segmenting selected sequence shallow shared shrinkage sigir signal singer single sixth skounakis some speech state statistical stochastic study support symposium systems table tagging task taskar technology text theocharous theory third time tishby transducers transformation tree treebank tutorial twelfth uncertainty using variational vector venkatesh very visual wainwright weighted weiss west willsky with workshop xiaoxiang yedidia
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/170.pdf	36	Feature Subset Selection for Learning Preferences: A Case Study	aaai accuracy acknowledgements acquisition adaptive advances alonso alvarez anaheim animal approach arti aseava assess assessment association asturiana australian authors automated bahamonde barnhill based bayesian beef been bengio between bootstrap bound bovine branting breeders broos california cancer carcasses cation cattle chapados chapter cial ciencia classes classi clickthrough clouse cohen comparison comparisons computer conference connectionist continuous cross data dataset design discovery each edinburgh engines erences estimation evaluation examples expert extensions feat feature features feder fiechter food from fulness function functions gene goyache graepel grant guyon help herbrich human iaai iberamia icann icml ieee ijcai implement indicate industry information intel intelligence international irvine jective joachims john joint journal kernel kinds kira know kohavi large learning ledge ligence light like linear luaces machine machines making margin margins mcyt meat methods metric mining ministerio model morphological names national neural ninth nips noisy number obermayer optimizing order ordinal over pair paper part pena percentage performance point practical predicate preference preferences press proc proceedings processing procs products quality quevedo quinlan ranilla references regression regularization relevant rendell reported research results reunanen rogers royo saxena scale schuurmans science search selection sets sevilla shapire singapore singer southey spain spanish standard stanford statistical studies study subset support supported system systems table techniques technology tecnolog tesauro thank their theory things this training transactions trends tting under user using utgo validation valles vapnik variable vector weston wiley wise with workshop would wrappers
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/211.pdf	73	Sequential Skewing: An Improved Skewing Algorithm	academy algorithm alternative altos application arti better biology blake brannetti carbonell cation cesareni cestra chapter chess cial cient classi conference cortactin databases decision distinct domains erent family fowlkes francisco from games gene generating helmercitterich homology ijcai induction intel international joint journal kaufmann learning ligand ligands ligence lookahead machine members merz michalski mitchell molecular morgan national norton page predict preferences preferred procedures proceedings programs publishers quillam quinlan references repository rider sciences skewing sparks spot their tree trees
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/180.pdf	46	Kernel-based Discriminative Learning Algorithms for Lab eling Sequences, Trees, and Graphs	advances algorithms altun boosting cambridge collins conference convolution data discriminative empirical experiments explorations francisco functions hidden hofmann information international investigating johnson kaufmann kernels label language learning loss machine machines markov methods models morgan natural neural optimization parsing perceptron press proceedings processing references reranking rtner sequences seventeenth sigkdd structured support survey systems theory training tsochantaridis twentieth vector with
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/253.pdf	33	Entropy-Based Criterion in Categorical Clustering	absence algorithm analysis aspects axiom barbara based baulieu baxter berlin bock categorical cation celeux cients cikm class classi cluster clustering coef conceptual conference coolcat couto criteria data differences discrete dissimilarity eleventh entropy govaert journal latent models monash numerical oliver opitz presence probabilistic proceedings references report similarities springer systems technical university variant verlag
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/207.pdf	60	Locally Linear Metric Adaptation for Semi-Sup ervised Clustering	acapulco adaptive advances algorithms american analysis appear application approximation arti association australia auxiliary background balance banerjee based basis basu becker boston breast cambridge canada cancer cardie cation cheung chin cial city classi clustering computation computing conditional conference constrained constraints criteria data deformable density department diabetes dimensionality discriminant distance distributions domeniconi eighteenth embedding equivalence erent estimates evaluation figure flexible four friedman from fukunaga functions gaussian girosi gunopulos hastie heisterkamp hertz hillel hostetler housing idealized ieee index information instance intel international ionosphere iris iterative jective joint jordan jorization journal kamvar kaski kernel kernels klein knearest knowledge kwok label learning level ligence linear locally lowe machine making manning means methods metric mexico mixture models mooney most multidimensional nearest neighbor networks neural nineteenth nonlinear numbered obermayer optimization parametric pattern peng plants plots poggio press prior proceedings processing protein radial rand recognition reduction references relations report results rogers roweis russell saul scaling schroedl science section seeding semisupervised sets seventeenth shental shown side similarity sinkkonen sixteenth soybean space stanford statistical statistics sydney systems technical theory thrun tibshirani transactions tsang twentieth university using variablekernel visual wagsta washington webb weinshall williamstown wine with xing yeung zhang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/250.pdf	53	Learning to Cluster using Lo cal Neighb orho o d Structure	account acknowledgments advances algorithm allows almost also alternative analysis another appear application applications approaches area areas arti ashburner associated available bach based been belief believe belkin between biology built categories cation certain changes cheng cial class classi cluster clustering clusterings comm comments common computer computing concentrate concept concepts connection connectivity consortium corduneanu could criterion cuts data decoding denoising density dependent derived dimensionality discussion distance distributions driven embedding encourage establishes experiments explicitly expression extension extensions extrapolate factor focs foundations framework freedom freeman frey from fully functional further future gave gene general generalized generative genet geometric global good graph graphs groups have helpful here high idea ideas ieee image immediate implicitly include inference inferring information instance intel inter interesting introduced itself jaakkola jordan journal just kannan kaufman kout kschischang label labeled labels landscape langford learn learning ligence ligent linear local locally loeliger machine mackay making malik manifolds many markov matrices mceliece measure meila method methods metric might model models more morgan morris mouse natural naturally neighborhood neighboring neural nition nity niyogi nonlinear normalized occur only ontology order other paper partially pattern pearl points posterior potentials presented preservation previous probabilistic problem problems processing product propagation properties proposed providing pursuing quaid random rather reasoning recently reduction references regarding regularization related relevant remarks research results reviewers riemannian rosales roweis russell same saul science seen segmentation selection semi semisupervised several should side silva simple since some speci spectral statistical structure studied submitted successful supervised systems szummer tenenbaum than thank that their theory there these this tool transactions turbo type uncert under unlike unsupervised upon used using valuable vempala vetta viewpoint walks weiss which with within work worth xing yedidia zhang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/123.pdf	5	A MFoM Learning Approach to Robust Multiclass Multi-Label Text Categorization	academic achieve achieves acknowledgements adaptation adaptive adopted advantage algorithm algorithms also analysis annual anonymous applications applied approach approximation artificial assigns audio automated automatic averaging back bakiri based bayes bellegarda best better binary boostexter boosting both breiman call cases categories categorization category chua class classical classification classifier classifiers classify clearly codes comments commonly communities comparison comparisons computing conclusion contrast correcting corresponding data decision descent design dietterich discriminative document dodier each empirical enhanced error esann especially examination example expect experimental exploiting extension fahlman family figure friedman from full function furthermore generalized give given goes good hauptmann high icml ieee improvements increase increased indicate information instances intelligence interest introduced joachims john journal juang katagiri kluwer known label language large latent learning less lewis like limited linear logistic machine machines macro mann maximal meantime meanwhile merit method methods mfom micro modeling modified more mostly mozer multi multiclass natural negative networks number obtained olshen only optimizing other output over paper parameter pattern performance performances popular positive probabilistic problems proc processing propagation proposed publishers qibin recognition references regression research results retrieval reuters reviewers ringuette robust robustness routers rules sample samples scale schapire scheme score sebastiani semantic separately should shows sigir significantly similar simultaneously since singer sizes solving sons sophisticate speech speed statistic statistical stone structures study support surveys symposium system take task test text than thank that their them theory these third this thus train trained training trans trees uniform upon using valuable vapnik variation vector wadsworth watkins weston which whitney wilcoxon wiley with wolniewicz work would yang zhang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/256.pdf	71	SVM-Based Generalized Multiple-Instance Learning via Approximate Box Counting	advances agnostic algorithms amar andrews approximation arti axisparallel bioinf blake burges carlo carlson cation chapter cial clyne computer data databases dietterich dooly enumeration from genomic geometric goldman hofmann html identi information instance intelligence joachims journal karp keogh kernel kwek large lathrop learning lozano luby machine machines madras making merz methods mlearn mlrepository monte moriyama multiple multitransmembrane neural novel patterns perez periodic practical press problem problems processing properties proteins quasi real rectangles references repository research scale scholkopf sciences scott smola solving structural support system systems tsochantaridis using valued vector warr with zhang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/246.pdf	20	Bellman go es Relational	abstract acknowledgements address advancing algorithm algorithms also american analysis anonymous applications approximate april arti association assumptions authors background barto based because been belgium bellman benelearn bias boutilier buntine cambridge cannot cant cases chapter cheng cial cient combining comments complexity compress computational conclusions conference constraint contract contribute contribution converge could data dean decision decisiontheoretic depending developed dialogue dietterich driessens dynamic early ective employing enable environments eroski european examples experimental explanation explanationbased extending feature fern first flann foundations fruhwirth functions further gartner gaussian gearhart generalized generalizing givan given graph guestrin handling hanks have helpful highlighted hope icml ijcai implement improved induction inductive insights instance intel intelligence interesting into introduction involved iteration jersey journal kanodia kernels kersting know knowledge koller language lational learn learning lecoeuche lecture ledge leverage ligence like linguistics logic logical machine management markov martijn mdps methods models naacl necessity netherlands nienhuys nips nite north notes novel number only operator optimal order other otterlo paper pittsburgh planning plans policies policy practice press price princeton problem problems proc processes programme programming programs raedt ramon rebel redundancy references regression reinforcement reiter relational representation represented research reviewers rmdps rmed rules second shown signi simple some springer states statistical structural structures subumption successful such supported sutton symbolic thank that their theoretical theory therefore this though turn types understanding union university update upgrade used using value verlag view well will with wolf work workshop would yoon
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/107.pdf	8	A Pitfall and Solution in Multi-Class Feature Selection for Text Classification	acknowledge acknowledgments also analysis artificial attribute barnhill based bayes benchmarking better cancer categorization centroid class classification cluster comparative computer conference cora cycles data dataset datasets development discovery discrete distribution document electrotechnical empirical engineering european examination experimental extensive fawcett feature features forman francisco frank gene gratefully grobelnik guyon hall holmes ieee implementations improving informatics information inria institute intelligence international java joachims john journal karypis kaufmann knowledge kohavi laboratory learning machine machines made many massachusetts master methods metrics mining mladenic morgan multi naive open paper pedersen practical prepared principles project references relevant rennie research results retrieval reviewers science selection sequences sigir software source study subset support techniques technology text thanks their thesis this tools transactions unbalanced using vapnik vector weka weston with witten word wrappers yang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/230.pdf	15	Approximate Inference by Markov Chains on Union Spaces	advances algorithm also analysis aperiodic aperiodicity applications approximate arbitrary arti bayesian belief berkeley between biometrika book both bound bremaud bristol california carlo chain chains changing chapter cial circle combination comc complex computation computer conj convergence converges corresponding deptarment determination determines dimensional distribution division easily ectral eigen eigenvalue eigenvalues eigenvectors elds equations equilibrium ergocity ergodic finally freeman generalized gibbs green hadamard highly however implies inference information inside intel ipsen irreducibility irreducible iterates jordan journal jump kappen kernel latter leisink ligence linear markov mathematics matrix mcmc means method meyer mixing model monte must neural only onverges other periodic plex power processing product propagation published pulls queues rate reducible references remedied report research reversible rosen satisfy scaling science siam simulation speed springer stability stochastic structure structured subdominant switch systems technical than that there this trans transition uniform unique university until vector vergence weiss welling which will with yedidia
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/61.pdf	34	Estimating Replicability of Classifier Learning Exp eriments	addison advances algorithms approach approximate avoid based bayesian bengio between blake bouckaert calibrated california cation choosing classi comparing computation concrete continuous data databases dietterich discovery distributions error estimating francisco frank generalization graham icml implementations inference information irvine java john kaufmann knowledge knuth langley learning machine mateo mathematics merz mining morgan nadeau neural patashnik pitfalls practical press processing programs publishers quinlan recommended references repository salzberg statistical supervised systems techniques tests tools university wesley with witten
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/76.pdf	76	Support Vector Machine Learning for Interdep endent and Structured Output Spaces	ability according accuracy achieve achieves added algorithm algorithmic algorithms align alignment altun application approach approaches arbitrary awkward based best better between bounds cant cases chapelle class collins comparable complete complex computational computed computer conclusions conditional conjecture constraints context conventional convergence cornell correctly covers crammer cult data department dependency discriminative distribution elds elissee emnlp empirical erence erty estimation examples exceeds exible experiments faster feature features formulated formulations foundations free functions further furthermore gains gave general generalization generative gives grammar grammars guestrin handle hidden hofmann holloway icml implementation impossible including incorporating indeed input interdependent interesting into joachims johnson joint kernel kernels koller labeling language large learning least line linguistic linguistics london lose loss machine machines manning margin marginally markov maximum maxmargin mccallum mcnemar method methods might model models more mostly multi naive natural networks nips note number often only optimization outperforming output outputs over overlapping pairs parameter parsing pcfg perceptron pereira practice predicting press probabilistic problems processing promising property proposed prove random range rather references remains report representations research resulting results roughly royal scaling scholkopf schuetze science score scores segmenting sequence sequences setting show shows signi simple since singer sixth small solve solving sons spaces speci sped statistical structured substantially supervised support table taskar technical terms test that theory this time total tractable train training tree tsochantaridis twice university used using values vapnik vector verify very watkins weighted were weston which while wide wiley with working workshop would zero
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/177.pdf	31	Efficient Hierarchical MCMC for Policy Search	advances american andrieu annealing applications approximation arti association bayesian biometrika calculations carlo chain chains chemical cial computing conference doucet equation evolutionary fast francisco freitas function gelatt godsill gradient hastings inference information intel introduction jordan journal kaufmann kernel kirkpatrick large learning liang ligence machine machines mansour markov mcallester mcmc mdps method methods metropolis mixture models monte morgan neal neural optimization parameter pegasus physics policy pomdps press probabilistic proceedings processing publishers real references regression reinforcement report rosenbluth sampling science scienti search sequential simulated singh springer state statistical strategies sutton systems technical teller their toronto uncertainty university using vecchi verlag vermaak with wong
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/113.pdf	9	A Spatio-temp oral Extension to Isomap Nonlinear Dimension Reduction	advances aldridge ambrose askew bengio bluethmann brand british burridge canada charting chiu clustering cohen collide collision columbia conference data detection diftler discovery eigenmaps envi exact extensions humanoid ieee information intel interactive international isomap keogh know large ledge ligent lonardi lovchik magruder manifold manocha mining motifs nasa neural nips paiement ponamgi probabilistic processing references rehnmark robonaut sample scale series sigkdd space spectral system systems time vancouver vincent washington
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/215.pdf	17	Automated Hierarchical Mixtures of Probabilistic Principal Comp onent Analyzers	akaike algorithm analysis archive articial assessing automatic bayesian biernacki bishop cation celeux clustering completed components conf control data dempster from govaert hierarchical identi ieee incomplete information integrated intel international laird latent ligence likelihood look machine maximum mixture model networks neural ninth pattern principal proceedings processing references rubin statistical systems tipping transactions variable variational visualization with
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/74.pdf	65	Online Learning of Conditionally I.I.D. Data	advances akademie algoet annual applying automation baum berlin bottou bound bounded cambridge chervonenkis clearly computation computational computer control convergent corol corollary dence devroye difference errn errors fact follows forecasting forty foundations from generalisation generalization german gives gyor haussler have ieee iinf information introduction isupp john kearns large lary learning lecun line lugosi machines made massachusetts maximal minimisation moreover morvai moscow nauka need neural nips nonparametric notations number online ordered pattern press probabilistic proceedings processing proof recognition references remote risk russian samples says scale science second series show size society sons springer statement stationary statistical summable symposium systems term that theorem theorie theory third time transactions translation using valid vapnik vazirani verlag vidyasagar vovk weakly wellcalibrated what which wiley yakowitz york zeichenerkennung
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/169.pdf	79	Testing the Significance of Attribute Interactions	about agreed agresti analysis analyzing annals approach arxiv assuming attribute average based bayes bayesian berger boggs bootstrap bratko breast categorical change chemical classes classi clustering combines comparison control cooper correlations could cross data decomposition dependencies distribution ecml edition entropy estimated ewsl expected explain figure fisher fold formulation frequency friedman from frustration function geiger german given goldszmidt good have higher horse http hypothesis improve including independence information informations interaction interactions intrinsic jakulin john journal kirkwood kononenko learning likelihood liquids loss lower lung lymph machine mathematical matsuda maximum mcgill mixture model monti multiple multivariate mutual naive nature network neyman nite order pazzani physical physics pkdd postop probability psychometrika quantifying radial references review reys rish science searching semi series small sons springer springerverlag squared statistical statistics test testing than that transmission upon using validation value values verlag vilalta visualizing wiley with
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/294.pdf	64	Nonparametric Classification with Polynomial MPMC Cascades	adaptive advances algorithm algorithms anderson annals approach approximation architecture arcing bahadur based bhattacharyya boost breiman cambridge cascade cation chang cjlin classi colt computational conf correlation covariance csie denver detection dimensional distributions erent estimation fahlman feedforward forests francisco freund friedman from function gallant ghaoui growth grudic high http ieee ijcai information intern international into joint jordan jority journal kaufmann kernels lanckriet lawrence learning lebiere libsvm machine mateo mathematical matrices matrix minimax morgan multivariate nadal nato network networks neural nonparametric normal outlier overview pena perceptron practical press prieto proc proceedings processing publishers random references research robust scholkopf smola spaces springerverlag statistics study systems technometrics theory trans version very with workshop
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/103.pdf	2	A Graphical Mo del for Protein Secondary Structure Prediction	acids alignment altschul application aurora barton biology blast burge capping complete conf database delcher function gapped gene generation genetics genomic goldberg helix human improve intel journal karlin kasif ligent lipman madden miller modelling molecular multiple networks nucleic prediction probabilistic proc programs protein proteins references research rose schae science search secondary sequence structure structures systems with zhang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/146.pdf	61	Lookahead-based Algorithms for Anytime Induction of Decision Trees	active algorithms arti baram based between blake blumer boddy bouckaert breiman brooks burges calibrated cation choice choosing cial classi constrained data databases dean deliberation discovery ehrenfeucht environments friedman haussler icml information intel know learning ledge letters ligence machine machines merz mining monterey occam olshen online pattern problem processing razor recognition references regression repository scheduling solving stone support tests time trees tutorial vector wadsworth warmuth yaniv
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/401.pdf	43	Interpolation-based Q-learning	accuracy adaptive addition aggregation algorithm algorithms also although anala appropriate approximation approximator approximators averagers bandwidths based basic basis being best cases chains combined computation concerns conclusions conditions construct control convergence decisions density derived discretization dynamic eligibility estimation even expansion extended feature function gordon guaranteed have high icml ijcai interpolation interpolative jaakkola jordan journal kaufmann kernel kernels knowledge large learning liebscher littman machine manner mansour markov match methods meyn mixing moore morgan munos neural nips optimal ormoneit point points press problems proc processes programming property qlearning quasi rates references regression reinforcement replacing require research resolution result results rigorous satisfy scale sense singh small soft solutions springer stability stable state statistics stochastic such sutton szepesv that they this traces tsitsiklis tweedie under value variable verlag when with ysis
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/198.pdf	80	Text Categorization with Many Redundant Features: Using Aggressive Feature Selection to Make SVMs Comp etitive with C4.5	advances algorithms analysis appear based bekkerman brank broad burges categorization cation chakrabarti cikm classi clustering cohen conference database datasets davidov department devices directories directory distributional duda dumais ecml electronic empirical extensive feature features fellbaum forman frayling gabrilovich generation grobelnik hart heckerman held herscovici hierarchical icml inductive inst interaction israel joachims john joshi journal kernel labeled large learning lexical linear maarek machine machines making many markovitch master methods metrics milic mladenic mobile models newbold parameterized pattern pennock personalized petruschka platt pocket practical press proc punera references relevant representations research sahami scale scene schoelkopf selection sigir smola sons structure study support technion technology text thesis topics vector wide wiley with wordnet words workshop world
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/96.pdf	23	Co-EM Supp ort Vector Learning	additional advances algorithm algorithms area asymptotic baluja bennett blum bradley brefeld categorization cation cirelo classi cohen collins combining computational computers conference continuum coop costs cozman curve data dempster denis discrimination eled empirical endent entity ervised european evaluation example examples face freeman from geib ghani gilleron icml ieee improvement incomplete information international journal kernel labeled laird language laurent learning likelihood machine machines mathematical maximum methods mitchell mixture modeling models multiclass named natural neural nonsup orientation ositive outcome pattern press probabilistic proceedings processing programming provided recognition references royal rubin semisup singer society statistical supp support systems text theory tommasi training transactions under unlab unlabeled unsup vector with workshop wysotzki
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/130.pdf	40	Gradient LASSO for feature selection	absolute accelerating acknowledgments adaptive additive advances algorithm anal analysis angle annals applic application approach approximation area asymptotics atomic australia australian bakin barron bartlett basis baxter bayes behavior boosting bounds building cambridge cant canu center chair chen classi cohn combining comments complex computer computing consistency constraint convergence data decisiontheoretic decomposition decreases department descent deviance deviances dissertation doctoral donoho dropped early ects efron equivalence estimators fast feature ferris figure force frean freund friedman fruitful function functional functions generalization gradient grafting grandvalet grant greedy gunn hastie hilbert hypotheses ieee important imposing improvements incremetal inform information jection johnstone jones journal kandola kearns kernels kimeldorf klein knight korea kosef lacker large larger lasso lassotype learnign learning least lemma library like likelihood line logistic lokhorst lugosi machine madison main margin mason mass math methods mining model modelling most national near network neural numerical optimum osborne outcomes parameters part perkins plus preg presentation presnell press problem problems processing proposed pursuing pursuit rates references regression regularized report represents research results reviewers ridge risk saunders schapire scholkopf schuurmans sciences scienti selection seoul shrinkage siam sigmoidal signi simple simulation slightly slowly smola solla solve some space sparse speed spline squares stage statist statistical statistics structural superpositions supported system systems tchebych technical techniques than thank that theiler then theory this those through tibshirani training trans turlach typical universal university unknown variable vayatis venables very view voelker wahba while whose wisconsin with worth would zhang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/13.pdf	41	Hyperplane Margin Classifiers on the Multinomial Manifold	advances algorithms also amari american apply approach asymptotic based bridson cation cencov classi comprehensive conference construct curvature curved decision developed differential diffusion dissertation doctoral documents dortmund exponential fisher foundations generalization geometrical geometry gous hall have here hofmann hyperplane iger inference information international introduction joachims john kass kernels lafferty language learning lebanon machine manifold manifolds many margin mathematical mathematics maximum methods metric models multinomial nagaoka natural neural nonpositive optimal perish point points presented processing properties publish rather references regression related retrieval riemannian rules science series simplex society sons spaces spherical spivak springer stanford statistical studies subfamilies subfamily summary text than theory treating treats under university using view voss wiley with work works
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/414.pdf	48	Learning First-order Rules from Data with Multiple Parts: Applications on Mining Chemical Comp ound Data	activity arti axis based chemical chevaleyre cial conf data dietterich discovery european feature finn flach framework from induction inductive instance intel international kaufmann kernels king kowalczyk lathrop learning ligence logic lozano machine morgan muggleton multi multiple order page parallel perez pharmacophore problem proc progol programming rectangles references relating rtner rules smola solving springer srinivasan sternberg structure study system using with zucker
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/379.pdf	50	Learning Low Dimensional Predictive Representations	aaai acting addresses advances against algorithm allowing also annual another arti based baum bayesian berkeley best better cally cases cassandra center cial cient circumstances compare computation computer concerns conclusion conference convincing could data decision department determining developed dimensional dimensionality discovery discrete domains easy eighteenth especially even exact example experiment experimental experiments extension fast found from further future general good have hidden hmms horizon icml ijcai important inadequate incremental induction information intelligence international involve izadi jaeger joint jong kaelbling large learn learning limited line littman long machine markov mccallum measure merging method methods model models modern more national neural nips nonlinear observable omohundro only operator optimal optimally order outcomes outperform parameters pardoe partially perception performance performed planning policy precup predicting prediction predictions predictive problem proceedings processes processing produce provide pruning psrs range real reduced references reinforcement report representation representations research restricted rochester rudary science selective separate series should shown simple simply singh situations smaller speci state stochastic stolcke stone street successfully such sutton systems technical test tests than that then these thesis thirteenth time tpsr tpsrs trained training tree twelfth twentieth uncertainty under university users welch where whether which will with work world would zhang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/264.pdf	1	A Fast Iterative Algorithm for Fisher Discriminant using Heterogeneous Kernels	academic adults advances afss alberta alder algorithm alternating analysis applications approach asscociation asymptomatic august avenue bach bartlett bennet berkeley bezdek boosting boston california cambridge cation cherkassky citations classi colonoscopy colorectal comput computational computer computing concepts conference convergence coupled cplex cristianini data diego discovery discriminant division edition edmonton ematical embrechts ensemble evgeniou fast finite fisher framework francisco from fukunaga fung fuzzy gebauer geetanjali generalized ghaoui hamers hathaway heterogeneous hill html http hung ieee ilog incline information institute international introduction issue john joint jordan journal july kernel kernels know lagrangian lanckriet large learning least ledge leemans letters lkopf machine machinery machines mangasarian margin mark math matha mathematics matrix mcgraw mcquaid method methods mika minimal mining mitchell mlearn mlrepository models momma moor mulier muller murphy nature nejm neoplasia networks neural neurocomputing nevada newton nips nite nonlinear notes optimization optimizationtechnical optimizer paral parmeterised pattern philadelphia poggio pontil press proceeding proceedings processing prog programming proximal recognition references regularization report repository research science screen second semide sequential shawe sher siam signal smooth some sons special springer squares ssvm statistical steinauer support suykens systems taylor technical techreports theory tomographic tsch university using vandewalle vapnik vector verlag village virtual wall weston wiley wisc wisconsin with york zhang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/354.pdf	37	Feature selection, L1 vs. L2 regularization, and rotational invariance	above absorbed according adopted algorithm also amongst another anthony anything applications argument arti assume assumption attain auer bartlett based basis because been begin best blum bound bounded bounding brief cally cambridge case cation certain change cial class classes classi combining complexity computation computing concept consider considering constant constraints continuous convention convergence correct covering covers cross decision denote details dimensions distribution does domain drawn dual earlier ehrenfeucht either endix ensure entirely equal equality equation equations ered error esti every everywhere example examples exist fact false features finally first found foundations from function functions further general generality generalization generalizations give given good greater hand haussler have hence here highly hold holder holds identical ieee implies imply important include independent indicai induced inequality information inner input inputs intelligence interest into introductions invariant ipping kearns kivinen known label labels langley large learning least lect like linear lipschitz logarithmic logistic logloss loop loss lower machine main mate matrix measured minimizing misclassi model more most must nearly necessarily necessary needed network networks neural nition nitions norm norms notation notice number numbers obtain omitted only optimization order original orthogonal other ously output over pand parameter parameterized parameters part particular pattern percep performance points pollard poly polynomial polynomially positive predictions predicts press previ probability problem problems proof properties proved putting quantities range readable recall recalling references referring regression regulariza relevant remaining replacing result resulting results right rotated rotationally same sample sampling satis second section select selected selection separators sets setting show showed shown shows side signs similarly simply since single size small smallest solving some space speci special standard stated statement step straightforward strang such summarize summary supz target test than that then theorem theoretic theoretical theory there therefore this thus tion together trained training transactions treat true type under uniform uniformly university upper used uses using valiant validation value values vapnik vector version viewed warmuth weights well were where which whose will with without worse would written zhang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/79.pdf	55	Learning to Track 3D Human Motion from Silhouettes	agarwal algorithm analysis applications approximation articulated athitsos based bayesian belongie bishop black blake body bombay brand bregler camera chapter cipolla cluttered computer conf context contexts correspondances cyclic darrell databases dhome distributions dynamical estimating estimation estimator european exponential fast features filtering freeman from grauman guibas gurations hand hashing hastie howe human hyperplane ieee image implicit inferring information intelligence intelligent invariant inverse jump jurie kinematic kinematics learned learning leventon linear local lowe maccormick machine malik maps matching metric model models monocular mori motion networks neural object objects ormoneit oxford parameter parts pattern pavlovic people piecewise point pose press probabilistic processes processing puppetry puzicha real recognition reconstruction references regression rehg relevance research robots rubner scale schaal sclaroff sensitive shadow shakhnarovich shape sidenbladh sigal silhouettes single sminchisescu souza sparse statistical stenger structure switching synthesis systems taylor template thayananthan time tipping tomasi torr tracking trans tree triggs twists uncalibrated university using vector video vijayakumar viola vision williams with without
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/91.pdf	39	Generalized Low Rank Approximations of Matrices	achlioptas aggarwal algebra approximate approximations barbara berry brie california castelli clustering computation conference csvd data decomposition deerwester dimensional dimensionality dumais ects engineering fast furnas harshman high ieee indexing information intelligent know landauer latent ledge linear matrix mcsherry pods proceedings rank reduction references retrieval review santa search searches seman siam similarity singular space stoc thomasian transactions using value
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/271.pdf	13	Adaptive Cognitive Orthotics: Combining Reinforcement Learning and Constraint-Based Temp oral Reasoning	activities adaptive addition allow along also annual approach architecture arti assistance assistive assn autominder autonomous average barto before begin both brown cambridge certain changes choice cial cient cognitive colbry college combination complex computational conclusion conference considered consistent constraint continued converge convergence conversational cult currently daily data days dechter default delayed deployed deploying deployment detail details developed dialogue dietterich disjunctive dissertation doctoral each environment evaluation even example experience experiment experiments extend feasibility fiechter figure finally forgotten frequently from functions generalizing gervasio goker have having here however human hypothesize identify impairment include individual instead integrated intel intelligent interact interaction interesting international introduction involves issue ject jective jectory joint journal kearns king kirsch langley large latter lead learn learning length lience ligence likely linguistics litman long lopresti machine made making management margins mccarthy meeting meiri memory method mihailidis more most move much must networks neuropsychological njfun note only optimal optimizing orthotic other pearl peintner people personalized pineau plan planning plans policy pollack possible press problem problems proceedings produce ramakrishnan random rapid real reasonable reasoning recommendations references rehabilitation reinforcement reminder reminders required research result results return rewards robotics robots rogers scheduling series several shop short should showed simple simulated simulation singh solution some spoken starting state studies study sutton system systems takes techniques technology temporal term text that this thompson those thrun time tsamardinos turn types useful user users variety walker watkins were when whether which will with work would zhang
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/195.pdf	83	Towards Tight Bounds for Rule Learning	aaai abstain abstaining abstinence accuracy accurate adopted adtrees advances algorithm algorithms allow allowing along also alternating annu applications applying arti asia averaging bagged bartlett based bayesian before bene between blake bound bounded bounds breiman calculated cambridge cases certain cial close cohen colt combinations compares comprehensible comput computation computational conf conference consensus considerations covering cross darmstadt data databases dataset datasets decision default dependent depending designed difference discovery discussion dissertation doctoral does easily effective empirical empirically ensemble ensembles error estimated experiment experiments extract fact fast favorably feature features figure fith forests framework frank frequently from further generalization generating global golea halfspaces hardcopy helps holmes hoos hope implementations important improved increase indicates induction indurkhya information informative instead intelligence international introduction japkowicz java kaufmann kearns kirkby knowledge known kramer kterm learner learning level lightweight lines list littlestone local machine majority making mansour marchand mason matching matter mcallester merz methods mining model models moreover morgan most narrowing national near neural noise noted optimal optimization optimizing over overall paci pakdd paper parameter particular performance pfahringer possible practical practice prediction predictions predictive preliminary presented press prior proc proceedings processing proved provide random ratio reasonably references relative repository ruckert rule rules schemes search seen sets setting shah shawe should showed shows simple singer sixteenth size smaller sokolova standard state statistics stochastic systems taylor techniques tenfold than that their theorem theoretical theoretically theory these this thus tightest tools trees uncertain used validation values variants varied various vazirani very warmuth ways weighted weiss which white with without witten work workshop
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/314.pdf	18	Bayesian Haplotyp e Inference via the Dirichlet Pro cess	american analysis annals association bayesian bioinformatics biology blocks cient clark computational daly density diploid escobar eskin estimation evolution exco ferguson frequencies from gabriel genetic genetics genome halperin haplotype high human inference inferences journal karp likelihood lipase lipoprotein maximum mixtures molecular nature nonparametric nucleotide perfect phylogeny population problems reconstruction references resolution science sequence slatkin some statistical statistics structure using variation west
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/161.pdf	66	Online and Batch Learning of Pseudo-Metrics	additive adjustment advances aggressive algebric algorithm algorithms analysis annual application applications average brain buckley cation censor chapman claderon classi clustering comp component computational computations conf convergence cover crammer dekel distance document duda eccv edition eigenvalue euro evaluating fast golub hall hart herbster hertz hopkins ieee information international john jordan kandola kernels learning length loan london machine macqueen margins math matrix means metric minimum mitra model models multidimensional nearest neighbor neural neurocomputing normalization online optimization organization oxford parallel partitions passive pattern pavel perceptron pivoted press probabilistic problem proc processing psychological references relevant reprinted retrieval review rosenblatt russell scaling shalev shawetaylor shental shwartz sideinformation singer singhal statist statistical storage stork systems theory transactions uneven university vapnik variance vision weinshall wiley wilkinson with xing zaragoza zenios
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/221.pdf	85	Unifying Collab orative and Content-Based Filtering	advances algorithms analysis arti balabanovic based basu billsus borchers breese cation cial classi claypool cohen collaborative collabrorative combining communications conference content contentbased crammer empiricial gokhale goldberg good heckerman herlocker hirsh information intel international kardie konstan learning ligence ltering lters machine miranda murnikov national netes neural newspaper nichols online pazzani pranking predictive proceedings processing ranking recommendation recommender references riedl sartin sarwar schafer shoham sigir singer social systems tapestry terry uncertainty using weave with workshop
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/392.pdf	56	Learning with Non-Positive Kernels	advanced algorithm alpay analysis applications appliquees approach aspects azizov bogn bottom column combination computational conference conjugate control convex data dawson deviation dissertation doctoral edition emerging engl estimate estimation experiments feature figure gaussian generalized gradient haasdonk hanke hansen hassibi herbrich hilbertian hyperkernels inde industrial inner institut international interpretation inverse iokhvidov john kailath kernel kernels kugler large learning lectures left linear lkopf longman machine march mary math mathematics mean mendelson methods metric middle models monographs national nips nite nonlinear notes observational operators ovari pitman posed positive press princeton problems proceedings product quadratic random references regularization representer reprint reproducing research results right rockafellar rouen sayed scale schur sciences scienti series shows siam sigmoid sinc smola some sons space spaces spline springer standard statistical study subdualities subspaces surveys svms system technical texts theorem theoretical theories theory training translated tutorials type ucla univ unpublished using vapnik verlag verse wahba wiley williamson with
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/206.pdf	25	Convergence of Synchronous Reinforcement Learning with Linear Function Approximation	according algorithms analysis applications applied approximation arbitrary assume athena automatic baird bertsekas case ccording choice choose cient combinations conclusions condition conditions construct control convergence convergent counterexample denotes depicted discretetime divergence diverges ecause edition either elaydi entry equations equiva equivalent erence erences erent factored following framework frontiers function general given gohberg greenbaum gure have icml ieee initial interscience introduction invariant iteration iterative koller lancaster learning least lent linear ludyk machine mathematics matrices matrix mdps merke methods must necessary neurodynamic nips nonzero numerical only optimality parr policy predict press problems programming prove proved publication references reinforcement residual results rodman rules schoknecht scienti seen siam solving spanned special springer squares stability subspace subspaces such sutton synchronous systems temporal that then theorem therefore this thus time transactions tsitsiklis update used values variant vector vieweg violated violates where which wiley with yields
http://kingman.cs.ualberta.ca/_banff04/icml/pages/papers/205.pdf	22	Boosting Margin Based Distance Functions for Clustering	academic adjustment advances alche algorithm ambroise analysis animal application applied averaged background bartlett based bengio blake blatt boosting bottom bottou cardie cases cation classes classi clustering coherence color comparing complete component computation computer computing conf conference constrained constraint constraints correct cumulative cuts data database databases datapoints dataset datasets dempster dence diego distance distboost document domany dtboost duda eccv edition effectiveness enhancing equivalence euclidean explanation figure francisco freund from fukunaga functions gaussian gdalyahu gradient grandvalet granular graphs grouping haffner hart hertz hilel hillel ieee image images improved incomplete index information instance intelligence international john jordan jrssb june kamvar kaufmann klein knowledge laird learnign learning lecun level likelihood linkage machine madison magnet making malik manning margin marginboost maximum means merz methods metric miller mixture mnist model models morgan most multimedia neighbor neighbors neural normalized number organization over pass pattern pavel perceptual predictions press prior proc proceedings processing purity rated realizations recognition references relations relevant repository results retrieval rogers rubin russell same schapire schroedl segmentation self semi shental shown side singer sons space statistical stochastic stork subset supervised systems transactions using vectors video vision voting wagstaff ward weinshall were werman wiley wiseman with xing zabih