Learning to Align Sequences: A Maximum-Margin Approach

Size: px
Start display at page:

Download "Learning to Align Sequences: A Maximum-Margin Approach"

Transcription

1 Learnng to Algn Sequences: A Maxmum-Margn Approach Thorsten Joachms Department of Computer Scence Cornell Unversty Ithaca, NY 4853 tj@cs.cornell.edu August 28, 2003 Abstract We propose a dscrmnatve method for learnng the parameters (e.g. cost of substtutons, deletons, nsertons) of lnear sequence algnment models from tranng examples. Whle the resultng tranng problem leads to an optmzaton problem wth an exponental number of constrants, we present a smple algorthm that fnds an arbtrarly close approxmaton after consderng only a subset of the constrants that s lnear n the number of tranng examples and polynomal n the length of the sequences. We also evaluate emprcally that the method effectvely learns good parameter values whle beng computatonally feasble. Introducton Methods for sequence algnment are common tools for analyzng sequence data rangng from bologcal applcatons [3] to natural language processng []. They can be thought of as measures of smlarty between sequences where the smlarty score s the result of a dscrete optmzaton problem that s typcally solved va dynamc programmng. Whle the dynamc programmng algorthm determnes the general noton of smlarty (e.g. local algnment vs. global algnment), any such smlarty measure requres specfc parameter values before t s fully specfed. Examples of such parameter values are the costs for substtutng one sequence elements for another, as well as costs for deletons and nsertons. These parameter values greatly determne how well the measure works for a partcular task. In ths paper we tackle the problem of nferrng the parameter values from tranng data. Our goal s to fnd parameter values so that the resultng smlarty measure best reflects the desred noton of smlarty. We consder tranng data where we have examples of smlar and dssmlar sequences. Instead of assumng a generatve model of sequence algnment (e.g. [9]), we take a dscrmnatve approach to tranng. In partcular, we am to fnd the set of parameter values that corresponds to the best smlarty

2 s : s 2 : a: m m s m m m m Fgure : Example of a local sequence algnment. measure a gven algnment algorthm can represent. Takng a large-margn approach, we show that we can solve the resultng tranng problem effcently for a large class of algnment algorthms that mplement a lnear scorng functon. Whle the resultng optmzaton problems have exponentally many constrants, our algorthm fnds an arbtrarly good approxmaton after consderng only a subset of constrants that scales polynomally wth the length of the sequences and lnearly wth the number of tranng examples. We emprcally and theoretcally analyze the scalng of the algorthm and show that the learned smlarty score performs well on test data. 2 Sequence Algnment Sequence algnment computes a smlarty score for two (or more) sequences s and s 2 from an alphabet ±=f; ::; ffg. An algnment a s a sequence of operatons that transforms one sequence nto the other. In global algnment, the whole sequence s transformed. In local algnment, only an arbtrarly szed subsequence s algned. Commonly used algnment operatons are match (m), substtuton (s), deleton (d) and nserton (). An example of a local algnment s gven n Fgure. In the example, there are 6 matches, substtuton, and nserton/deleton. Wth each operaton there s an assocated cost/reward. Assumng a reward of 3 for match, a cost of for substtuton, and a cost of 2 for nserton/deleton, the total algnment score D ~w (s ;s 2 ;a) n the example s 5. The optmal algnment a Λ s the one that maxmzes the score for a gven cost model. More generally, we consder algnment algorthms that optmze a lnear scorng functon D ~w (s ;s 2 ;a)= ~w T ff(s ;s 2 ;a) where ff(s ;s 2 ;a) s some functon that generates features based on an algnment a (e.g. # of substtutons, # of deletons) and w s a gven cost vector. ff(s ;s 2 ;a) depends on the partcular algnment algorthm and can be any feature vector. Fndng the optmal algnment corresponds to the followng optmzaton problem D ~w (s ;s 2 )=max a [D ~w (s ;s 2 ;a)] = max a ~w T ff(s ;s 2 ;a) Λ : () Ths type of problem s typcally solved va dynamc programmng. In the followng we consder local algnment va the Smth/Waterman algorthm [0]. However, the results can be extended to any algnment algorthm that optmzes a lnear scorng functon and that solves () globally optmally. Ths also holds for other structures besdes sequences. 2

3 3 Inverse Sequence Algnment Inverse sequence algnment s the problem of usng tranng data to learn the parameters ~w of an algnment model and algorthm so that the resultng smlarty measure D ~w (s ;s 2 ) best represents the desred noton of smlarty on new data. Whle prevous approaches to ths problem exst [5, ], they are lmted to specal cases and very small numbers of parameters. We wll present an algorthm that apples to any lnear algnment model wth no restrcton on the functon or number of parameters. We assume the followng model, for whch the notaton s nspred by proten algnment. In proten algnment the goal s a smlarty measure so that homologous proten sequences are smlar to a natve sequence, but also so that non-homologous (.e. decoy) sequences are not smlar to the natve sequence. We assume that examples are generated..d. accordng to a dstrbuton P (s N ;s H ;S D ). s N s the natve sequence, s H the homologous sequence, and S D s a set of decoy sequences s D ; :::; s D d. The goal s a smlarty measure so that natve sequence s N and homolog s H are more smlar than the natve sequence s N and any decoy s Dj,.e. D ~w (s N ;s H ) >D ~w (s N ;s Dj ): (2) The goal of learnng can be formulated n two reasonable ways based on two dfferent loss functons. In the frst settng, the goal s to fnd the cost parameters ~w that mnmze the probablty Err P ( ~w) that the smlarty wth any decoy sequence D ~w(s N ;s Dj ) s hgher than the smlarty wth the homolog sequence D ~w (s N ;s H ). Err P ( ~w) = Z max D ~w (s N ;s Dj ) >D ~w (s N ;s H ) dp (s N ;s H ;S D ) (3) s D j 2S D The functon (:) returns, f ts argument s false, 0 otherwse. In the second settng, the goal s less ambtous. We do not want a homolog that s necessarly more smlar than all decoys. Instead, we merely want to have the homolog sequence ranked hghly, but not necessarly on top of the rankng. In partcular, we could optmze the average rank of the homolog. Ths s equvalent to the followng knd of error rate. Z X ErrP f D ( ~w) = ~w (s N ;s Dj ) >D ~w (s N ;s ) H dp (s N ;s H ;S D ) (4) s D j 2S D In the followng, we consder only the frst type of error Err P ( ~w) for concseness. 4 A Maxmum-Margn Approach to Learnng the Cost Parameters The dstrbuton P (s D ;s H ;S D ) s unknown. However, we have a tranng sample S from P (s N ;s H ;S D ). It conssts of natve sequences s N ; :::; sn n, homolog sequences s H ; :::; sh n, and a set of decoy sequences S D ; :::; SD n for each natve sn ;:::;sn n. As a smplfyng assumpton, we assume that between natve and homolog sequences the algnment a NH ; :::; a NH n of maxmum score s known 2. The goal s to fnd an optmal 2 For proten algnment, for example, ths could be generated va structural algnment. 3

4 ~w so that the error rate Err P ( ~w) s low. Frst, we consder the case where there exsts a ~w such that the error on the tranng set Err S ( ~w) = n X = ψ max s D j 2S D! D ~w (s N ;sdj ) <D ~w (s N ;sh ) s zero. Snce we assume a scorng functon that s lnear n the parameters D ~w (s ;s 2 ;a)=~w T ff(s ;s 2 ;a); (6) the condton of zero tranng error can be wrtten as a set of lnear nequalty constrants. There s one constrant for each combnaton of natve sequence s N, decoy sequence s Dj, and possble algnment a of s N nto s Dj. 2 S D 8a : D ~w (s N ;sdj ;a) <D ~w(s N ;sh ;anh ) ::: (7) n 2 SD n 8a : D ~w(s N n ;sdj n ;a) <D ~w(s N n ;sh n ;anh ) Ths approach of wrtng the tranng problem as a lnear system follows the method n [8] proposed for the specal case of global algnment wthout free nsertons/deletons. However, for the general case n Equaton (7) the number of constrants s exponental, snce the number of algnments a between s N and s Dj s exponental n the length of s N and s Dj. Unlke the restrcted case n [8], standard optmzaton algorthms cannot handle ths sze problem. To overcome ths lmtaton, n Secton 5 we wll propose an algorthm that explots the specal structure of Equaton (7) so that t needs to examne only a subset that s at most lnear n the number of tranng examples n. If the set of nequaltes n Equaton (7) s feasble, there wll typcally be more than one soluton ~w Λ. To specfy a unque soluton, we select the ~w for whch the smlarty of the homolog D ~w (s N n ;sh n ;pnh ) s unformly most dfferent from the smlarty max D s j 2S D D w (s N ;sd ) of the best algnment wth any decoy. Ths corresponds to the maxmum-margn prncple employed n Support Vector Machnes [2]. Denotng the margn by ff and restrctng the L 2 norm of ~w to make the problem well-posed, ths leads to the followng optmzaton problem. max ~w (5) 2 ff2 (8) 2 S D 8a : D ~w(s N ;sdj ;a)» D ~w(s N ;sh ;anh ) ff (9) ::: (0) n 2 Sn D 8a : D ~w (s N n ;sdj n ;a)» D ~w(s N n ;sh n ;anh ) ff () jj ~wjj = (2) Due to the lnearty of the smlarty functon (6), the length of ~w s a free varable and we can fx t to =ff. Substtutng for ff and rearrangng leads to the equvalent optmzaton problem mn ~w 2 ~wt ~w (3) 4

5 2 S D 8a : ff(s N ;sh ;anh ) ff(s N ;sdj ;a) ~w (4) ::: (5) ;a) n 2 Sn D 8a : ff(s N n ;s H n ;anh ) ff(s N n ;sdj n ~w (6) Snce ths quadratc program (QP) has a postve-defnte objectve functon and (feasble) lnear constrants, t s strctly convex. Ths means t has a unque global mnmum and no local mnma [4]. The constrants are smlar to the ordnal regresson approach n [6] and t has a structure smlar to the Rankng SVM descrbed n [7] for nformaton retreval. However, the number of constrants s much larger. To allow errors n the tranng set, we ntroduce slack varables ο [2]. Correspondng to the error measure Err P ( ~w) we have one slack varable for each natve sequence. Ths s dfferent from a normal classfcaton or regresson SVM, where there s a dfferent slack varable for each constrant. The slacks enter the objectve functon accordng to a trade-off parameter C. For smplcty, we consder only the case where the slacks enter the objectve functon squared. nx mn ~w; ο ~ 2 ~wt ~w + C ο 2 (7) = 2 S D 8a : ff(s N ;sh ;anh ) ff(s N ;sdj ;a) ~w ο (8) ::: (9) ;a) n 2 SD n 8a : ff(s N n ;s H n ;anh ) ff(s N n ;sdj n ~w ο n (20) P n Note that = ο2 s an upper bound on the tranng error Err S ( ~w). Therefore, the algorthm mnmzes the tranng error Err S ( ~w) whle maxmzng margn. 5 Tranng Algorthm Due to the exponental number of constrants, standard optmzaton software cannot handle the number of constrants resultng from problems of nterestng sze. However, by explotng the specal structure of the problem, we propose the followng algorthm that fnds the soluton of (7)-(20) after examnng only a small number of constrants. The algorthm proceeds by greedly addng constrants from (8)-(20) to a workng set K. The algorthm stops, when all constrants n (8)-(20) are fulflled up to a precson of ffl. 5

6 Input: natve sequences s N ; :::; sn n, homolog sequences sh ; :::; sh n, algnments anh ;:::;a NH n, sets of decoy sequences S D ;:::;SD n, tolerated approxmaton error ffl 0. K = ;, ~w =0, ~ ο =0 repeat ffl K org = K ffl for from to n untl(k = K org ) Output: ~w for j from to js D j va dynamc program- h Λ fnd a Λ = argmax a ~w T ff(s N ;sdj ;a) mng ~w T (ff(s N ;sh ;anh ) ff(s N ;sdj Λ f ~w T (ff(s N ;sh ;anh ) ff(s N ;sdj ;a Λ )) < ο ffl n o K = K [ ;a Λ )) ο P solve QP ( ~w; ο)=argmn ~ ~w; ο ~ 2 ~wt n ~w + C K. = ο2 subject to The followng two theorems show that the algorthm returns the correct soluton and that t stops after O(n) teratons through the repeat-loop. Theorem (CORRECTNESS) The algorthm returns an approxmaton to (7)-(20) that has an objectve value not hgher than the soluton of (7)-(20), and that fulflls all constrants up to a precson of ffl. Forffl =0, the algorthm returns the soluton ( ~w Λ ; ~ ο Λ ) of (7)-(20). Proof Let ~w Λ and ο Λ be the soluton of (7)-(20). Snce the algorthm solves the QP on a subset P of the constrants n each P teraton, t must return a soluton ~w wth 2 ~wt ~w + n C = ο2» 2 ~wλt ~w Λ n + C = ολ2. Ths follows from the fact that decreasng the feasble regon cannot lead to a lower mnmum. It s left to show that the algorthm does not termnate before all constrants (8)- (20) are fulflled up to precson ffl. In the fnal teraton, the algorthm fnds the hghest scorng algnment a Λ for each decoy and checks whether the constrant ~w T (ff(s N ;sh ;anh ) ff(s N ;sdj ;a Λ )) < ο ffl s fulflled. Three cases can occur: Frst, the constrant can be volated and the algorthm wll not termnate yet. Second, t s already n K and s fulflled by constructon. Thrd, the constrant s not n K but fulflled anyway and t s not added. If the constrant s fulflled, the constrants for all other algnments nto ths decoy are fulflled as well, snce we checked the constrant for whch the margn was smallest for the gven ~w. It follows that the algorthm termnates only f all constrants are fulflled up to precson ffl. Smlarly, t can be shown that for ffl>0 the algorthm fnds a soluton so that the KKT-condtons of the Wolfe Dual of (7)-(20) are fulflled up to a precson of ffl. We omt the proof for brevty. 6

7 Theorem 2 (TERMINATION) The algorthm stops after addng at most 2V R 2 (2) ffl 2 constrants to the set K. V s the mnmum of (7)-(20) and R 2 s a constant bounded by the maxmum of (ff(s N ;sh ;anh ) ff(s N ;sdj ;a)) 2 + 2C. Proof The frst part of the proof s to show that the objectve value ncreases by some constant wth every constrant that P s added to K. Denote wth V k the soluton V k = P ( ~w k Λ ; ο ~ k Λ ) = mn ~w; ο ~ 2 ~wt n ~w + C = ο2 subject to K k after addng k constrants. Ths prmal optmzaton problem can be transformed nto an equvalent problem of the form V k = P ( ~w k 0Λ )=mn ~w 0 2 ~w0t ~w 0 subject to Kk 0, where each constrant has the form ~w T ~x wth p ~x =(ff(s N ;sh ;anh ) ff(s N ;sdj ;a Λ ); 0; :::;0;= 2C;0;:::;0). Its correspondng Wolfe dual s D(~ff Λ k )=max ~ff 0 P k = ff 2 P k = P k j= ff ff j ~x ~x j. At the soluton D(~ff Λ k ) = P ( ~w0λ k ) = P ( ~wλ k ; ο ~ k Λ ) = V k and for every feasble pont D(~ff)» P ( ~w; ο). ~ Prmal and dual are P P connected va ~w 0Λ k = = ffλ ~x. Addng a constrant to the dual wth ~w 0ΛT k ~x k+ = = ffλ ~x ~x k+» ffl means extendng the dual to kx kx kx kx D k+ (~ff Λ k+ )= max ff ff ff j ~x ~x j + ff k+ ff k+ ff ~x ~x k+ ~ff k ff2 k+ ~x2 k+ = = j= D k (~ff Λ k )+ max ff k+ ff k+ ff k+ 0 kx = = ff Λ ~x ~x k+ 2 ff2 k+ ~x2 k+ D k (~ff Λ k )+ max ff k+ ff k+ ( ffl) ff k+ 0 2 ff2 k+ ~x2 k+ Solvng the remanng scalar optmzaton problem over ff k+ shows that ff Λ k+ 0 and that V k+ V k + ffl2 2R. 2 Snce the algorthm only adds constrants that are volated by the current soluton by more than ffl, after addng k max = 2VR2 ffl constrants the soluton 2 V kmax over the subset K kmax s at least V kmax V 0 + 2VR2 ffl 2 ffl 2 2R 2 =0+V. Any addtonal constrant that s volated by more than ffl would lead to a mnmum that s larger than V. Snce the mnmum over a subset of constrants can only be smaller than the mnmum over all constrants, there cannot be any more constrants volated by more than ffl and the algorthm stops. The theorem drectly leads to the concluson that the maxmum number of constrants n K scales lnearly wth the number of tranng examples n, snce V can be upper bounded as V» C Λ n usng the feasble pont ~w =0and ~ ο =n (7)-(20). Furthermore, t scales only polynomally wth the length of the sequences, snce R s polynomal n the length of the sequences. Whle the number of constrants can potentally explode for small values of ffl, experence wth Support Vector Machnes for classfcaton showed that relatvely large 7

8 ThreeParam Tran ThreeParam Test IndvSubstCost Tran IndvSubstCost Test Average Tran/Test Error (delta) Number of Tranng Examples Fgure 2: Left: Tran and test error rates for the 3 and the 403 parameter model dependng on the number of tranng examples. Rght: Typcal learned substtuton matrx after 40 tranng examples for the 403-parameter model. values of ffl are suffcent wthout loss of generalzaton performance. We wll verfy the effcency and the predcton performance of the algorthm emprcally n the followng. 6 Experments To analyze the behavor of the algorthm under varyng condtons, we constructed a synthetc dataset accordng to the followng sequence and algnment model. The natve sequence and the decoys are generated by drawng randomly from a 20 letter alphabet ±=f; ::; 20g so that letter c 2 ± has probablty c=20. Each sequence has length 50, and there are 0 decoys per natve. To generate the homolog, we generate an algnment strng of length 30 consstng of 4 characters match, substtute, nsert, delete. For smplcty of llustraton, substtutons are always c! (c mod 20) +. Whle we experment wth several algnment models, we only report typcal results here where matches occur wth probablty 0:2, substtutons wth 0:4, nserton wth 0:2, deleton wth 0:2. The homolog s created by applyng the algnment strng to a randomly selected substrng of the natve. The shortenng of the sequences through nsertons and deletons s padded by addtonal random characters. Fgure 2 shows tranng and test error rates for two models dependng on the number of tranng examples averaged over 0 trals. The frst model has only 3 parameters ( match, substtute, nsert/delete ) and uses a unform substtuton matrx. The second model also learns the 400 parameters of the substtuton matrx, resultng n a total of 403 parameters. We chose C = 0:0 and ffl = 0:. The left-hand graph of Fgure 2 shows that for the 403-parameter model, the generalzaton error s hgh for small numbers of tranng examples, but quckly drops as the number of examples ncreases. The 3-parameter model cannot ft the data as well. Its tranng error starts out much hgher and tranng and test error essentally converge after only a few examples. The rght-hand graph of Fgure 2 shows the learned matrx of substtuton costs for the 403-parameter model. As desred, the elements of the matrx are close to zero except for the off-dagonal. Ths captures the substtuton model c! (c mod 20)+. Fgure 3 analyzes the effcency of the algorthm va the number of constrants that 8

9 ThreeParam IndvSubstCost ThreeParam IndvSubstCost Average Number of Constrants Average Number of Constrants Number of Tranng Examples Epslon Fgure 3: Number of constrants added to K dependng on the number of tranng examples (left) and the value of ffl (rght). If not stated otherwse, ffl =0:, C =0:0, and n =20. are added to K before convergence. The left-hand graph shows the scalng wth the number of tranng examples. As predcted by Theorem 2, the number of constrants grows (sub-)lnearly wth the number of examples. Furthermore, the actual number of constrants s small enough so that t can easly be handled by standard quadratc optmzaton software. The rght-hand graph shows how the number of constrants n the fnal K changes wth log(ffl). The observed scalng appears to be better than suggested by the upper bound n Theorem 2. A good value for ffl s 0:. We observed that larger values lead to worse predcton accuracy, whle smaller values decrease effcency whle not provdng further beneft. 7 Conclusons The paper presented a dscrmnatve learnng approach to nferrng the cost parameters of a lnear sequence algnment model from tranng data. We proposed an algorthm for solvng the resultng tranng problem and showed ts effcency both theoretcally and emprcally. Experments show that the algorthm can effectvely learn the algnment parameters on a synthetc task. We are currently applyng the algorthm to learnng algnment models for proten homology detecton. An open queston s whether t s possble to remove the assumpton that the algnment between natve and homolog s known whle mantanng the tractablty of the problem. Acknowledgments Many thanks to Ron Elber and Tamara Galor for ther nsghtful dscussons and ther contrbuton to the research that lead to ths paper. 9

10 References [] R. Barzlay and L. Lee. Bootstrappng lexcal choce va multple-sequence algnment. In Conference on Emprcal Methods n Natural Language Processng (EMNLP), [2] Cornna Cortes and Vladmr N. Vapnk. Support vector networks. Machne Learnng Journal, 20: , 995. [3] R. Durbn, S. Eddy, A. Krogh, and G. Mtchson. Bologcal Sequence Analyss. Cambrdge Unversty Press, 998. [4] P. E. Gll, W. Murray, and M. H. Wrght. Practcal Optmzaton. Academc Press, 98. [5] D. Gusfeld and P. Stellng. Parametrc and nverse-parametrc sequence algnment wth xparal. Methods n Enzymology, 266:48 494, 996. [6] R. Herbrch, T. Graepel, and K. Obermayer. Large margn rank boundares for ordnal regresson. In Advances n Large Margn Classfers, pages MIT Press, Cambrdge, MA, [7] T. Joachms. Optmzng search engnes usng clckthrough data. In Proceedngs of the ACM Conference on Knowledge Dscovery and Data Mnng (KDD), [8] J. Meller and R. Elber. Lnear programmng optmzaton and a double statstcal flter for proten threadng protocols. Protens Structure, Functon, and Genetcs, 45:24 26, 200. [9] S. E Rstad and P. N. Yanlos. Learnng strng edt dstance. IEEE Transactons on Pattern Recognton and Machne Intellgence, Vol. 20(5): , 998. [0] T. Smth and M. Waterman. Identfcaton of common molecular subsequences. Journal of Molecular Bology, 47:95 97, 98. [] Fangtng Sun, D. Fernandez-Baca, and We Yu. Inverse parametrc sequence algnment. In Internatonal Computng and Combnatorcs Conference (CO- COON), [2] V. Vapnk. Statstcal Learnng Theory. Wley, Chchester, GB,

Natural Language Processing and Information Retrieval

Natural Language Processing and Information Retrieval Natural Language Processng and Informaton Retreval Support Vector Machnes Alessandro Moschtt Department of nformaton and communcaton technology Unversty of Trento Emal: moschtt@ds.untn.t Summary Support

More information

Kernel Methods and SVMs Extension

Kernel Methods and SVMs Extension Kernel Methods and SVMs Extenson The purpose of ths document s to revew materal covered n Machne Learnng 1 Supervsed Learnng regardng support vector machnes (SVMs). Ths document also provdes a general

More information

Learning to Align Sequences: A Maximum-Margin Approach

Learning to Align Sequences: A Maximum-Margin Approach Learning to Align Sequences: A Maximum-Margin Approach Thorsten Joachims Tamara Galor Ron Elber Department of Computer Science Cornell University Ithaca, NY 14853 {tj,galor,ron}@cs.cornell.edu June 24,

More information

Lecture 10 Support Vector Machines II

Lecture 10 Support Vector Machines II Lecture 10 Support Vector Machnes II 22 February 2016 Taylor B. Arnold Yale Statstcs STAT 365/665 1/28 Notes: Problem 3 s posted and due ths upcomng Frday There was an early bug n the fake-test data; fxed

More information

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017 U.C. Berkeley CS94: Beyond Worst-Case Analyss Handout 4s Luca Trevsan September 5, 07 Summary of Lecture 4 In whch we ntroduce semdefnte programmng and apply t to Max Cut. Semdefnte Programmng Recall that

More information

Lecture Notes on Linear Regression

Lecture Notes on Linear Regression Lecture Notes on Lnear Regresson Feng L fl@sdueducn Shandong Unversty, Chna Lnear Regresson Problem In regresson problem, we am at predct a contnuous target value gven an nput feature vector We assume

More information

Support Vector Machines. Vibhav Gogate The University of Texas at dallas

Support Vector Machines. Vibhav Gogate The University of Texas at dallas Support Vector Machnes Vbhav Gogate he Unversty of exas at dallas What We have Learned So Far? 1. Decson rees. Naïve Bayes 3. Lnear Regresson 4. Logstc Regresson 5. Perceptron 6. Neural networks 7. K-Nearest

More information

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification E395 - Pattern Recognton Solutons to Introducton to Pattern Recognton, Chapter : Bayesan pattern classfcaton Preface Ths document s a soluton manual for selected exercses from Introducton to Pattern Recognton

More information

The Minimum Universal Cost Flow in an Infeasible Flow Network

The Minimum Universal Cost Flow in an Infeasible Flow Network Journal of Scences, Islamc Republc of Iran 17(2): 175-180 (2006) Unversty of Tehran, ISSN 1016-1104 http://jscencesutacr The Mnmum Unversal Cost Flow n an Infeasble Flow Network H Saleh Fathabad * M Bagheran

More information

Module 9. Lecture 6. Duality in Assignment Problems

Module 9. Lecture 6. Duality in Assignment Problems Module 9 1 Lecture 6 Dualty n Assgnment Problems In ths lecture we attempt to answer few other mportant questons posed n earler lecture for (AP) and see how some of them can be explaned through the concept

More information

MMA and GCMMA two methods for nonlinear optimization

MMA and GCMMA two methods for nonlinear optimization MMA and GCMMA two methods for nonlnear optmzaton Krster Svanberg Optmzaton and Systems Theory, KTH, Stockholm, Sweden. krlle@math.kth.se Ths note descrbes the algorthms used n the author s 2007 mplementatons

More information

Generalized Linear Methods

Generalized Linear Methods Generalzed Lnear Methods 1 Introducton In the Ensemble Methods the general dea s that usng a combnaton of several weak learner one could make a better learner. More formally, assume that we have a set

More information

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg prnceton unv. F 17 cos 521: Advanced Algorthm Desgn Lecture 7: LP Dualty Lecturer: Matt Wenberg Scrbe: LP Dualty s an extremely useful tool for analyzng structural propertes of lnear programs. Whle there

More information

Lecture 10 Support Vector Machines. Oct

Lecture 10 Support Vector Machines. Oct Lecture 10 Support Vector Machnes Oct - 20-2008 Lnear Separators Whch of the lnear separators s optmal? Concept of Margn Recall that n Perceptron, we learned that the convergence rate of the Perceptron

More information

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems Numercal Analyss by Dr. Anta Pal Assstant Professor Department of Mathematcs Natonal Insttute of Technology Durgapur Durgapur-713209 emal: anta.bue@gmal.com 1 . Chapter 5 Soluton of System of Lnear Equatons

More information

Chapter Newton s Method

Chapter Newton s Method Chapter 9. Newton s Method After readng ths chapter, you should be able to:. Understand how Newton s method s dfferent from the Golden Secton Search method. Understand how Newton s method works 3. Solve

More information

Assortment Optimization under MNL

Assortment Optimization under MNL Assortment Optmzaton under MNL Haotan Song Aprl 30, 2017 1 Introducton The assortment optmzaton problem ams to fnd the revenue-maxmzng assortment of products to offer when the prces of products are fxed.

More information

2.3 Nilpotent endomorphisms

2.3 Nilpotent endomorphisms s a block dagonal matrx, wth A Mat dm U (C) In fact, we can assume that B = B 1 B k, wth B an ordered bass of U, and that A = [f U ] B, where f U : U U s the restrcton of f to U 40 23 Nlpotent endomorphsms

More information

Linear Approximation with Regularization and Moving Least Squares

Linear Approximation with Regularization and Moving Least Squares Lnear Approxmaton wth Regularzaton and Movng Least Squares Igor Grešovn May 007 Revson 4.6 (Revson : March 004). 5 4 3 0.5 3 3.5 4 Contents: Lnear Fttng...4. Weghted Least Squares n Functon Approxmaton...

More information

COS 521: Advanced Algorithms Game Theory and Linear Programming

COS 521: Advanced Algorithms Game Theory and Linear Programming COS 521: Advanced Algorthms Game Theory and Lnear Programmng Moses Charkar February 27, 2013 In these notes, we ntroduce some basc concepts n game theory and lnear programmng (LP). We show a connecton

More information

Difference Equations

Difference Equations Dfference Equatons c Jan Vrbk 1 Bascs Suppose a sequence of numbers, say a 0,a 1,a,a 3,... s defned by a certan general relatonshp between, say, three consecutve values of the sequence, e.g. a + +3a +1

More information

Lecture 12: Classification

Lecture 12: Classification Lecture : Classfcaton g Dscrmnant functons g The optmal Bayes classfer g Quadratc classfers g Eucldean and Mahalanobs metrcs g K Nearest Neghbor Classfers Intellgent Sensor Systems Rcardo Guterrez-Osuna

More information

APPENDIX A Some Linear Algebra

APPENDIX A Some Linear Algebra APPENDIX A Some Lnear Algebra The collecton of m, n matrces A.1 Matrces a 1,1,..., a 1,n A = a m,1,..., a m,n wth real elements a,j s denoted by R m,n. If n = 1 then A s called a column vector. Smlarly,

More information

Support Vector Machines CS434

Support Vector Machines CS434 Support Vector Machnes CS434 Lnear Separators Many lnear separators exst that perfectly classfy all tranng examples Whch of the lnear separators s the best? + + + + + + + + + Intuton of Margn Consder ponts

More information

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009 College of Computer & Informaton Scence Fall 2009 Northeastern Unversty 20 October 2009 CS7880: Algorthmc Power Tools Scrbe: Jan Wen and Laura Poplawsk Lecture Outlne: Prmal-dual schema Network Desgn:

More information

On the Multicriteria Integer Network Flow Problem

On the Multicriteria Integer Network Flow Problem BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 5, No 2 Sofa 2005 On the Multcrtera Integer Network Flow Problem Vassl Vasslev, Marana Nkolova, Maryana Vassleva Insttute of

More information

Which Separator? Spring 1

Which Separator? Spring 1 Whch Separator? 6.034 - Sprng 1 Whch Separator? Mamze the margn to closest ponts 6.034 - Sprng Whch Separator? Mamze the margn to closest ponts 6.034 - Sprng 3 Margn of a pont " # y (w $ + b) proportonal

More information

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty Addtonal Codes usng Fnte Dfference Method Benamn Moll 1 HJB Equaton for Consumpton-Savng Problem Wthout Uncertanty Before consderng the case wth stochastc ncome n http://www.prnceton.edu/~moll/ HACTproect/HACT_Numercal_Appendx.pdf,

More information

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur Module 3 LOSSY IMAGE COMPRESSION SYSTEMS Verson ECE IIT, Kharagpur Lesson 6 Theory of Quantzaton Verson ECE IIT, Kharagpur Instructonal Objectves At the end of ths lesson, the students should be able to:

More information

On the Repeating Group Finding Problem

On the Repeating Group Finding Problem The 9th Workshop on Combnatoral Mathematcs and Computaton Theory On the Repeatng Group Fndng Problem Bo-Ren Kung, Wen-Hsen Chen, R.C.T Lee Graduate Insttute of Informaton Technology and Management Takmng

More information

Ensemble Methods: Boosting

Ensemble Methods: Boosting Ensemble Methods: Boostng Ncholas Ruozz Unversty of Texas at Dallas Based on the sldes of Vbhav Gogate and Rob Schapre Last Tme Varance reducton va baggng Generate new tranng data sets by samplng wth replacement

More information

Solutions to exam in SF1811 Optimization, Jan 14, 2015

Solutions to exam in SF1811 Optimization, Jan 14, 2015 Solutons to exam n SF8 Optmzaton, Jan 4, 25 3 3 O------O -4 \ / \ / The network: \/ where all lnks go from left to rght. /\ / \ / \ 6 O------O -5 2 4.(a) Let x = ( x 3, x 4, x 23, x 24 ) T, where the varable

More information

Support Vector Machines

Support Vector Machines Support Vector Machnes Konstantn Tretyakov (kt@ut.ee) MTAT.03.227 Machne Learnng So far Supervsed machne learnng Lnear models Least squares regresson Fsher s dscrmnant, Perceptron, Logstc model Non-lnear

More information

Support Vector Machines CS434

Support Vector Machines CS434 Support Vector Machnes CS434 Lnear Separators Many lnear separators exst that perfectly classfy all tranng examples Whch of the lnear separators s the best? Intuton of Margn Consder ponts A, B, and C We

More information

LECTURE 9 CANONICAL CORRELATION ANALYSIS

LECTURE 9 CANONICAL CORRELATION ANALYSIS LECURE 9 CANONICAL CORRELAION ANALYSIS Introducton he concept of canoncal correlaton arses when we want to quantfy the assocatons between two sets of varables. For example, suppose that the frst set of

More information

Maximal Margin Classifier

Maximal Margin Classifier CS81B/Stat41B: Advanced Topcs n Learnng & Decson Makng Mamal Margn Classfer Lecturer: Mchael Jordan Scrbes: Jana van Greunen Corrected verson - /1/004 1 References/Recommended Readng 1.1 Webstes www.kernel-machnes.org

More information

Pattern Classification

Pattern Classification Pattern Classfcaton All materals n these sldes ere taken from Pattern Classfcaton (nd ed) by R. O. Duda, P. E. Hart and D. G. Stork, John Wley & Sons, 000 th the permsson of the authors and the publsher

More information

LOW BIAS INTEGRATED PATH ESTIMATORS. James M. Calvin

LOW BIAS INTEGRATED PATH ESTIMATORS. James M. Calvin Proceedngs of the 007 Wnter Smulaton Conference S G Henderson, B Bller, M-H Hseh, J Shortle, J D Tew, and R R Barton, eds LOW BIAS INTEGRATED PATH ESTIMATORS James M Calvn Department of Computer Scence

More information

Problem Set 9 Solutions

Problem Set 9 Solutions Desgn and Analyss of Algorthms May 4, 2015 Massachusetts Insttute of Technology 6.046J/18.410J Profs. Erk Demane, Srn Devadas, and Nancy Lynch Problem Set 9 Solutons Problem Set 9 Solutons Ths problem

More information

Kristin P. Bennett. Rensselaer Polytechnic Institute

Kristin P. Bennett. Rensselaer Polytechnic Institute Support Vector Machnes and Other Kernel Methods Krstn P. Bennett Mathematcal Scences Department Rensselaer Polytechnc Insttute Support Vector Machnes (SVM) A methodology for nference based on Statstcal

More information

The Order Relation and Trace Inequalities for. Hermitian Operators

The Order Relation and Trace Inequalities for. Hermitian Operators Internatonal Mathematcal Forum, Vol 3, 08, no, 507-57 HIKARI Ltd, wwwm-hkarcom https://doorg/0988/mf088055 The Order Relaton and Trace Inequaltes for Hermtan Operators Y Huang School of Informaton Scence

More information

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 12 10/21/2013. Martingale Concentration Inequalities and Applications

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 12 10/21/2013. Martingale Concentration Inequalities and Applications MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.65/15.070J Fall 013 Lecture 1 10/1/013 Martngale Concentraton Inequaltes and Applcatons Content. 1. Exponental concentraton for martngales wth bounded ncrements.

More information

Homework Assignment 3 Due in class, Thursday October 15

Homework Assignment 3 Due in class, Thursday October 15 Homework Assgnment 3 Due n class, Thursday October 15 SDS 383C Statstcal Modelng I 1 Rdge regresson and Lasso 1. Get the Prostrate cancer data from http://statweb.stanford.edu/~tbs/elemstatlearn/ datasets/prostate.data.

More information

1 Convex Optimization

1 Convex Optimization Convex Optmzaton We wll consder convex optmzaton problems. Namely, mnmzaton problems where the objectve s convex (we assume no constrants for now). Such problems often arse n machne learnng. For example,

More information

Learning Theory: Lecture Notes

Learning Theory: Lecture Notes Learnng Theory: Lecture Notes Lecturer: Kamalka Chaudhur Scrbe: Qush Wang October 27, 2012 1 The Agnostc PAC Model Recall that one of the constrants of the PAC model s that the data dstrbuton has to be

More information

Lecture 3: Dual problems and Kernels

Lecture 3: Dual problems and Kernels Lecture 3: Dual problems and Kernels C4B Machne Learnng Hlary 211 A. Zsserman Prmal and dual forms Lnear separablty revsted Feature mappng Kernels for SVMs Kernel trck requrements radal bass functons SVM

More information

CSE 252C: Computer Vision III

CSE 252C: Computer Vision III CSE 252C: Computer Vson III Lecturer: Serge Belonge Scrbe: Catherne Wah LECTURE 15 Kernel Machnes 15.1. Kernels We wll study two methods based on a specal knd of functon k(x, y) called a kernel: Kernel

More information

Support Vector Machines

Support Vector Machines Support Vector Machnes Konstantn Tretyakov (kt@ut.ee) MTAT.03.227 Machne Learnng So far So far Supervsed machne learnng Lnear models Non-lnear models Unsupervsed machne learnng Generc scaffoldng So far

More information

CSC 411 / CSC D11 / CSC C11

CSC 411 / CSC D11 / CSC C11 18 Boostng s a general strategy for learnng classfers by combnng smpler ones. The dea of boostng s to take a weak classfer that s, any classfer that wll do at least slghtly better than chance and use t

More information

INF 5860 Machine learning for image classification. Lecture 3 : Image classification and regression part II Anne Solberg January 31, 2018

INF 5860 Machine learning for image classification. Lecture 3 : Image classification and regression part II Anne Solberg January 31, 2018 INF 5860 Machne learnng for mage classfcaton Lecture 3 : Image classfcaton and regresson part II Anne Solberg January 3, 08 Today s topcs Multclass logstc regresson and softma Regularzaton Image classfcaton

More information

Lecture 20: November 7

Lecture 20: November 7 0-725/36-725: Convex Optmzaton Fall 205 Lecturer: Ryan Tbshran Lecture 20: November 7 Scrbes: Varsha Chnnaobreddy, Joon Sk Km, Lngyao Zhang Note: LaTeX template courtesy of UC Berkeley EECS dept. Dsclamer:

More information

Online Classification: Perceptron and Winnow

Online Classification: Perceptron and Winnow E0 370 Statstcal Learnng Theory Lecture 18 Nov 8, 011 Onlne Classfcaton: Perceptron and Wnnow Lecturer: Shvan Agarwal Scrbe: Shvan Agarwal 1 Introducton In ths lecture we wll start to study the onlne learnng

More information

The Study of Teaching-learning-based Optimization Algorithm

The Study of Teaching-learning-based Optimization Algorithm Advanced Scence and Technology Letters Vol. (AST 06), pp.05- http://dx.do.org/0.57/astl.06. The Study of Teachng-learnng-based Optmzaton Algorthm u Sun, Yan fu, Lele Kong, Haolang Q,, Helongang Insttute

More information

Single-Facility Scheduling over Long Time Horizons by Logic-based Benders Decomposition

Single-Facility Scheduling over Long Time Horizons by Logic-based Benders Decomposition Sngle-Faclty Schedulng over Long Tme Horzons by Logc-based Benders Decomposton Elvn Coban and J. N. Hooker Tepper School of Busness, Carnege Mellon Unversty ecoban@andrew.cmu.edu, john@hooker.tepper.cmu.edu

More information

Kernels in Support Vector Machines. Based on lectures of Martin Law, University of Michigan

Kernels in Support Vector Machines. Based on lectures of Martin Law, University of Michigan Kernels n Support Vector Machnes Based on lectures of Martn Law, Unversty of Mchgan Non Lnear separable problems AND OR NOT() The XOR problem cannot be solved wth a perceptron. XOR Per Lug Martell - Systems

More information

A PROBABILITY-DRIVEN SEARCH ALGORITHM FOR SOLVING MULTI-OBJECTIVE OPTIMIZATION PROBLEMS

A PROBABILITY-DRIVEN SEARCH ALGORITHM FOR SOLVING MULTI-OBJECTIVE OPTIMIZATION PROBLEMS HCMC Unversty of Pedagogy Thong Nguyen Huu et al. A PROBABILITY-DRIVEN SEARCH ALGORITHM FOR SOLVING MULTI-OBJECTIVE OPTIMIZATION PROBLEMS Thong Nguyen Huu and Hao Tran Van Department of mathematcs-nformaton,

More information

Time-Varying Systems and Computations Lecture 6

Time-Varying Systems and Computations Lecture 6 Tme-Varyng Systems and Computatons Lecture 6 Klaus Depold 14. Januar 2014 The Kalman Flter The Kalman estmaton flter attempts to estmate the actual state of an unknown dscrete dynamcal system, gven nosy

More information

10-701/ Machine Learning, Fall 2005 Homework 3

10-701/ Machine Learning, Fall 2005 Homework 3 10-701/15-781 Machne Learnng, Fall 2005 Homework 3 Out: 10/20/05 Due: begnnng of the class 11/01/05 Instructons Contact questons-10701@autonlaborg for queston Problem 1 Regresson and Cross-valdaton [40

More information

Economics 101. Lecture 4 - Equilibrium and Efficiency

Economics 101. Lecture 4 - Equilibrium and Efficiency Economcs 0 Lecture 4 - Equlbrum and Effcency Intro As dscussed n the prevous lecture, we wll now move from an envronment where we looed at consumers mang decsons n solaton to analyzng economes full of

More information

Durban Watson for Testing the Lack-of-Fit of Polynomial Regression Models without Replications

Durban Watson for Testing the Lack-of-Fit of Polynomial Regression Models without Replications Durban Watson for Testng the Lack-of-Ft of Polynomal Regresson Models wthout Replcatons Ruba A. Alyaf, Maha A. Omar, Abdullah A. Al-Shha ralyaf@ksu.edu.sa, maomar@ksu.edu.sa, aalshha@ksu.edu.sa Department

More information

MAXIMUM A POSTERIORI TRANSDUCTION

MAXIMUM A POSTERIORI TRANSDUCTION MAXIMUM A POSTERIORI TRANSDUCTION LI-WEI WANG, JU-FU FENG School of Mathematcal Scences, Peng Unversty, Bejng, 0087, Chna Center for Informaton Scences, Peng Unversty, Bejng, 0087, Chna E-MIAL: {wanglw,

More information

Feature Selection: Part 1

Feature Selection: Part 1 CSE 546: Machne Learnng Lecture 5 Feature Selecton: Part 1 Instructor: Sham Kakade 1 Regresson n the hgh dmensonal settng How do we learn when the number of features d s greater than the sample sze n?

More information

Linear Classification, SVMs and Nearest Neighbors

Linear Classification, SVMs and Nearest Neighbors 1 CSE 473 Lecture 25 (Chapter 18) Lnear Classfcaton, SVMs and Nearest Neghbors CSE AI faculty + Chrs Bshop, Dan Klen, Stuart Russell, Andrew Moore Motvaton: Face Detecton How do we buld a classfer to dstngush

More information

Appendix for Causal Interaction in Factorial Experiments: Application to Conjoint Analysis

Appendix for Causal Interaction in Factorial Experiments: Application to Conjoint Analysis A Appendx for Causal Interacton n Factoral Experments: Applcaton to Conjont Analyss Mathematcal Appendx: Proofs of Theorems A. Lemmas Below, we descrbe all the lemmas, whch are used to prove the man theorems

More information

VARIATION OF CONSTANT SUM CONSTRAINT FOR INTEGER MODEL WITH NON UNIFORM VARIABLES

VARIATION OF CONSTANT SUM CONSTRAINT FOR INTEGER MODEL WITH NON UNIFORM VARIABLES VARIATION OF CONSTANT SUM CONSTRAINT FOR INTEGER MODEL WITH NON UNIFORM VARIABLES BÂRZĂ, Slvu Faculty of Mathematcs-Informatcs Spru Haret Unversty barza_slvu@yahoo.com Abstract Ths paper wants to contnue

More information

6.854J / J Advanced Algorithms Fall 2008

6.854J / J Advanced Algorithms Fall 2008 MIT OpenCourseWare http://ocw.mt.edu 6.854J / 18.415J Advanced Algorthms Fall 2008 For nformaton about ctng these materals or our Terms of Use, vst: http://ocw.mt.edu/terms. 18.415/6.854 Advanced Algorthms

More information

3.1 ML and Empirical Distribution

3.1 ML and Empirical Distribution 67577 Intro. to Machne Learnng Fall semester, 2008/9 Lecture 3: Maxmum Lkelhood/ Maxmum Entropy Dualty Lecturer: Amnon Shashua Scrbe: Amnon Shashua 1 In the prevous lecture we defned the prncple of Maxmum

More information

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming EEL 6266 Power System Operaton and Control Chapter 3 Economc Dspatch Usng Dynamc Programmng Pecewse Lnear Cost Functons Common practce many utltes prefer to represent ther generator cost functons as sngle-

More information

Support Vector Machines. Jie Tang Knowledge Engineering Group Department of Computer Science and Technology Tsinghua University 2012

Support Vector Machines. Jie Tang Knowledge Engineering Group Department of Computer Science and Technology Tsinghua University 2012 Support Vector Machnes Je Tang Knowledge Engneerng Group Department of Computer Scence and Technology Tsnghua Unversty 2012 1 Outlne What s a Support Vector Machne? Solvng SVMs Kernel Trcks 2 What s a

More information

For now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results.

For now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results. Neural Networks : Dervaton compled by Alvn Wan from Professor Jtendra Malk s lecture Ths type of computaton s called deep learnng and s the most popular method for many problems, such as computer vson

More information

Linear Feature Engineering 11

Linear Feature Engineering 11 Lnear Feature Engneerng 11 2 Least-Squares 2.1 Smple least-squares Consder the followng dataset. We have a bunch of nputs x and correspondng outputs y. The partcular values n ths dataset are x y 0.23 0.19

More information

CIS526: Machine Learning Lecture 3 (Sept 16, 2003) Linear Regression. Preparation help: Xiaoying Huang. x 1 θ 1 output... θ M x M

CIS526: Machine Learning Lecture 3 (Sept 16, 2003) Linear Regression. Preparation help: Xiaoying Huang. x 1 θ 1 output... θ M x M CIS56: achne Learnng Lecture 3 (Sept 6, 003) Preparaton help: Xaoyng Huang Lnear Regresson Lnear regresson can be represented by a functonal form: f(; θ) = θ 0 0 +θ + + θ = θ = 0 ote: 0 s a dummy attrbute

More information

Finding Dense Subgraphs in G(n, 1/2)

Finding Dense Subgraphs in G(n, 1/2) Fndng Dense Subgraphs n Gn, 1/ Atsh Das Sarma 1, Amt Deshpande, and Rav Kannan 1 Georga Insttute of Technology,atsh@cc.gatech.edu Mcrosoft Research-Bangalore,amtdesh,annan@mcrosoft.com Abstract. Fndng

More information

Yong Joon Ryang. 1. Introduction Consider the multicommodity transportation problem with convex quadratic cost function. 1 2 (x x0 ) T Q(x x 0 )

Yong Joon Ryang. 1. Introduction Consider the multicommodity transportation problem with convex quadratic cost function. 1 2 (x x0 ) T Q(x x 0 ) Kangweon-Kyungk Math. Jour. 4 1996), No. 1, pp. 7 16 AN ITERATIVE ROW-ACTION METHOD FOR MULTICOMMODITY TRANSPORTATION PROBLEMS Yong Joon Ryang Abstract. The optmzaton problems wth quadratc constrants often

More information

The Geometry of Logit and Probit

The Geometry of Logit and Probit The Geometry of Logt and Probt Ths short note s meant as a supplement to Chapters and 3 of Spatal Models of Parlamentary Votng and the notaton and reference to fgures n the text below s to those two chapters.

More information

Case A. P k = Ni ( 2L i k 1 ) + (# big cells) 10d 2 P k.

Case A. P k = Ni ( 2L i k 1 ) + (# big cells) 10d 2 P k. THE CELLULAR METHOD In ths lecture, we ntroduce the cellular method as an approach to ncdence geometry theorems lke the Szemeréd-Trotter theorem. The method was ntroduced n the paper Combnatoral complexty

More information

a b a In case b 0, a being divisible by b is the same as to say that

a b a In case b 0, a being divisible by b is the same as to say that Secton 6.2 Dvsblty among the ntegers An nteger a ε s dvsble by b ε f there s an nteger c ε such that a = bc. Note that s dvsble by any nteger b, snce = b. On the other hand, a s dvsble by only f a = :

More information

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix Lectures - Week 4 Matrx norms, Condtonng, Vector Spaces, Lnear Independence, Spannng sets and Bass, Null space and Range of a Matrx Matrx Norms Now we turn to assocatng a number to each matrx. We could

More information

Some modelling aspects for the Matlab implementation of MMA

Some modelling aspects for the Matlab implementation of MMA Some modellng aspects for the Matlab mplementaton of MMA Krster Svanberg krlle@math.kth.se Optmzaton and Systems Theory Department of Mathematcs KTH, SE 10044 Stockholm September 2004 1. Consdered optmzaton

More information

CS 3710: Visual Recognition Classification and Detection. Adriana Kovashka Department of Computer Science January 13, 2015

CS 3710: Visual Recognition Classification and Detection. Adriana Kovashka Department of Computer Science January 13, 2015 CS 3710: Vsual Recognton Classfcaton and Detecton Adrana Kovashka Department of Computer Scence January 13, 2015 Plan for Today Vsual recognton bascs part 2: Classfcaton and detecton Adrana s research

More information

Newton s Method for One - Dimensional Optimization - Theory

Newton s Method for One - Dimensional Optimization - Theory Numercal Methods Newton s Method for One - Dmensonal Optmzaton - Theory For more detals on ths topc Go to Clck on Keyword Clck on Newton s Method for One- Dmensonal Optmzaton You are free to Share to copy,

More information

APPROXIMATE PRICES OF BASKET AND ASIAN OPTIONS DUPONT OLIVIER. Premia 14

APPROXIMATE PRICES OF BASKET AND ASIAN OPTIONS DUPONT OLIVIER. Premia 14 APPROXIMAE PRICES OF BASKE AND ASIAN OPIONS DUPON OLIVIER Prema 14 Contents Introducton 1 1. Framewor 1 1.1. Baset optons 1.. Asan optons. Computng the prce 3. Lower bound 3.1. Closed formula for the prce

More information

Support Vector Machines

Support Vector Machines CS 2750: Machne Learnng Support Vector Machnes Prof. Adrana Kovashka Unversty of Pttsburgh February 17, 2016 Announcement Homework 2 deadlne s now 2/29 We ll have covered everythng you need today or at

More information

Solutions HW #2. minimize. Ax = b. Give the dual problem, and make the implicit equality constraints explicit. Solution.

Solutions HW #2. minimize. Ax = b. Give the dual problem, and make the implicit equality constraints explicit. Solution. Solutons HW #2 Dual of general LP. Fnd the dual functon of the LP mnmze subject to c T x Gx h Ax = b. Gve the dual problem, and make the mplct equalty constrants explct. Soluton. 1. The Lagrangan s L(x,

More information

Lagrange Multipliers Kernel Trick

Lagrange Multipliers Kernel Trick Lagrange Multplers Kernel Trck Ncholas Ruozz Unversty of Texas at Dallas Based roughly on the sldes of Davd Sontag General Optmzaton A mathematcal detour, we ll come back to SVMs soon! subject to: f x

More information

P R. Lecture 4. Theory and Applications of Pattern Recognition. Dept. of Electrical and Computer Engineering /

P R. Lecture 4. Theory and Applications of Pattern Recognition. Dept. of Electrical and Computer Engineering / Theory and Applcatons of Pattern Recognton 003, Rob Polkar, Rowan Unversty, Glassboro, NJ Lecture 4 Bayes Classfcaton Rule Dept. of Electrcal and Computer Engneerng 0909.40.0 / 0909.504.04 Theory & Applcatons

More information

Singular Value Decomposition: Theory and Applications

Singular Value Decomposition: Theory and Applications Sngular Value Decomposton: Theory and Applcatons Danel Khashab Sprng 2015 Last Update: March 2, 2015 1 Introducton A = UDV where columns of U and V are orthonormal and matrx D s dagonal wth postve real

More information

Supporting Information

Supporting Information Supportng Informaton The neural network f n Eq. 1 s gven by: f x l = ReLU W atom x l + b atom, 2 where ReLU s the element-wse rectfed lnear unt, 21.e., ReLUx = max0, x, W atom R d d s the weght matrx to

More information

Computational Biology Lecture 8: Substitution matrices Saad Mneimneh

Computational Biology Lecture 8: Substitution matrices Saad Mneimneh Computatonal Bology Lecture 8: Substtuton matrces Saad Mnemneh As we have ntroduced last tme, smple scorng schemes lke + or a match, - or a msmatch and -2 or a gap are not justable bologcally, especally

More information

A Bayes Algorithm for the Multitask Pattern Recognition Problem Direct Approach

A Bayes Algorithm for the Multitask Pattern Recognition Problem Direct Approach A Bayes Algorthm for the Multtask Pattern Recognton Problem Drect Approach Edward Puchala Wroclaw Unversty of Technology, Char of Systems and Computer etworks, Wybrzeze Wyspanskego 7, 50-370 Wroclaw, Poland

More information

Complete subgraphs in multipartite graphs

Complete subgraphs in multipartite graphs Complete subgraphs n multpartte graphs FLORIAN PFENDER Unverstät Rostock, Insttut für Mathematk D-18057 Rostock, Germany Floran.Pfender@un-rostock.de Abstract Turán s Theorem states that every graph G

More information

Logistic Regression. CAP 5610: Machine Learning Instructor: Guo-Jun QI

Logistic Regression. CAP 5610: Machine Learning Instructor: Guo-Jun QI Logstc Regresson CAP 561: achne Learnng Instructor: Guo-Jun QI Bayes Classfer: A Generatve model odel the posteror dstrbuton P(Y X) Estmate class-condtonal dstrbuton P(X Y) for each Y Estmate pror dstrbuton

More information

Structured Perceptrons & Structural SVMs

Structured Perceptrons & Structural SVMs Structured Perceptrons Structural SVMs 4/6/27 CS 59: Advanced Topcs n Machne Learnng Recall: Sequence Predcton Input: x = (x,,x M ) Predct: y = (y,,y M ) Each y one of L labels. x = Fsh Sleep y = (N, V)

More information

18-660: Numerical Methods for Engineering Design and Optimization

18-660: Numerical Methods for Engineering Design and Optimization 8-66: Numercal Methods for Engneerng Desgn and Optmzaton n L Department of EE arnege Mellon Unversty Pttsburgh, PA 53 Slde Overve lassfcaton Support vector machne Regularzaton Slde lassfcaton Predct categorcal

More information

Report on Image warping

Report on Image warping Report on Image warpng Xuan Ne, Dec. 20, 2004 Ths document summarzed the algorthms of our mage warpng soluton for further study, and there s a detaled descrpton about the mplementaton of these algorthms.

More information

A 2D Bounded Linear Program (H,c) 2D Linear Programming

A 2D Bounded Linear Program (H,c) 2D Linear Programming A 2D Bounded Lnear Program (H,c) h 3 v h 8 h 5 c h 4 h h 6 h 7 h 2 2D Lnear Programmng C s a polygonal regon, the ntersecton of n halfplanes. (H, c) s nfeasble, as C s empty. Feasble regon C s unbounded

More information

Classification as a Regression Problem

Classification as a Regression Problem Target varable y C C, C,, ; Classfcaton as a Regresson Problem { }, 3 L C K To treat classfcaton as a regresson problem we should transform the target y nto numercal values; The choce of numercal class

More information

Speeding up Computation of Scalar Multiplication in Elliptic Curve Cryptosystem

Speeding up Computation of Scalar Multiplication in Elliptic Curve Cryptosystem H.K. Pathak et. al. / (IJCSE) Internatonal Journal on Computer Scence and Engneerng Speedng up Computaton of Scalar Multplcaton n Ellptc Curve Cryptosystem H. K. Pathak Manju Sangh S.o.S n Computer scence

More information

Mean Field / Variational Approximations

Mean Field / Variational Approximations Mean Feld / Varatonal Appromatons resented by Jose Nuñez 0/24/05 Outlne Introducton Mean Feld Appromaton Structured Mean Feld Weghted Mean Feld Varatonal Methods Introducton roblem: We have dstrbuton but

More information

The exam is closed book, closed notes except your one-page cheat sheet.

The exam is closed book, closed notes except your one-page cheat sheet. CS 89 Fall 206 Introducton to Machne Learnng Fnal Do not open the exam before you are nstructed to do so The exam s closed book, closed notes except your one-page cheat sheet Usage of electronc devces

More information