Extending Relevance Model for Relevance Feedback
|
|
- Warren Lamb
- 5 years ago
- Views:
Transcription
1 Extendng Relevance Model for Relevance Feedback Le Zhao, Chenmn Lang and Jame Callan Language Technologes Insttute School of Computer Scence Carnege Mellon Unversty {lezhao, chenmnl, Abstract Relevance feedback s the retreval task where the system s gven not only an nformaton need, but also some relevance judgement nformaton, usually from users feedback for an ntal result lst by the system. Wth dfferent amount of feedback nformaton avalable, the optmal feedback strategy mght be very dfferent. In TREC Relevance Feedback task, the system s gven dfferent sets of feedback nformaton from 1 relevant document to over 40 judgements wth at least 3 relevant. Thus, n ths work, we try to develop a feedback algorthm that works well on all levels of feedback by extendng the relevance model for pseudo relevance feedback to nclude judged relevant documents when scorng feedback terms. Wthn these dfferent levels of feedback, t s more dffcult for the feedback algorthm to perform well when gven mnmal amount of feedback. Experments show that our algorthm performs robustly n those dffcult cases. 1 Introducton Relevance feedback s the retreval task where the system s gven not only a user query, but also user feedback on some of the top ranked results. Feedback gves the retreval system a chance to mprove ts results by explotng the extra nformaton through more elaborate technques. Ths can be helpful n cases where the users want as many relevant results as possble. Expermentng wth and falure analyses on feedback algorthms also provdes us a chance to understand what are mssng n users keyword queres and how to mprove these queres. In the followng, we frst ntroduce the relevance feedback task defned by ths year s track, the bass ad hoc retreval algorthm. In Secton 2, we layout the basc model for combnng pseudo relevance feedback and true relevance feedback n query expanson. One dffculty n dong that s properly and effectvely balancng term weghts from the two sources. We gve our soluton. We show experments n Secton 3 and conclude. 1.1 Task defnton Wth ths year s relevance feedback experments, the retreval system wll be evaluated on dfferent number of feedback documents. Specfcally, for each query, set A s the baselne wthout any feedback nformaton. Set B allows the system to use 1 relevant document as postve feedback; set C allows 3 relevant and 3 non-relevant; set D uses 10 judged documents (a superset of set C), and set E provdes all judged documents to the retreval algorthm, varyng from 40 to 100s of judged documents. 1.2 Pseudo relevance feedback as bass algorthm Snce dfferent numbers of feedback documents are possble, especally, when only gven a small number of judged documents, any relevance feedback algorthm that only look at the few judged documents can have dffculty provdng relable expanson queres. Ths s because fewer feedback documents means larger varatons of term appearances and expanson weghts. The approach we propose here, s to leverage the small amount of feedback nformaton wth some pseudo relevance feedback, so that the larger number of feedback documents provde lower varance estmates of the feedback terms, stablzng the performance of the feedback algorthm.
2 The motvaton for usng the partcular pseudo relevance feedback method proposed n (Lavrenko and Croft 2001) s that the best runs on the two prevous TREC 2006 and 2007 Terabyte tracks, (Metzler et al 2005) and (L and Yan 2006), both used ths partcular pseudo relevance feedback method. Snce the relevance feedback task uses the same collecton and smlar set of queres, we set the baselne retreval method to be the same as n the two prevous top runs a dependency model to acheve hgher top precson and a pseudo relevance feedback step followng. Because of ts relevance to ths work, we show the pseudo relevance feedback model mplemented n the Indr search engne: Pr ( I) = D topn Pr ( DPI ) ( DPD ) ( ) PI ( ) Relevance_model_query: #weght(w 1 r 1 w 2 r 2.. w n r n ), where, w = P(r I). Fnal_query: #weght( (1 - w) Relevance_model_query w Orgnal_query ) Here, r s a concept term, whch could refer to any term or phrase (we restrct ourselves to only keywords), D refers to a feedback document, and I means the nformaton need whch s expressed by the query. P(r I) s a multnomal dstrbuton over all concepts, of how deally the relevant class should look lke. P(I D) s approxmated wth the probablty for a document to generate the query, whch s just the language modelng verson of the relevance score. As P(I) s constant w.r.t. r, P(r I) s easly calculated f we assume P(D) s unform. Top weghted terms are selected for query expanson and weghted exactly as ther weghts, so that the expanson query ranks documents accordng to ther KL-dvergence to the relevance model P(r I). The fnal query s just nterpolatng the expanson query wth the orgnal query. The orgnal queres can be the keyword query ttles provded by TREC or the dependency model phrase queres whch we used as bass algorthm. Now we ntroduce our relevance feedback algorthm. 2 Relevance feedback algorthm The desgn of a feedback algorthm would be easy f the system s gven complete judgments for top N results, just use the judged documents as tranng samples, use smple term features and tran a bnary classfer for relevant or not, see for example (Zhu, Zhao and Callan 2007). Ths wll be especally effectve f N s large. However, n many cases the system does not have such a luxury. In the current track, 3 out of the 4 feedback sets have N as low as 1, 6 or 10. Thus, n ths secton we propose a unfed pseudo-relevance and true relevance feedback method based on the relevance model formalsm (Lavrenko and Croft 2001). From equaton (1) usng Bayes rule, we get the orgnal form of equaton (1), Pr ( I) = Pr ( DPD ) ( I) (2) D feedback For the feedback documents that have relevance judgments,.e. D D T, we trust the judgments, thus we set P(D I) = Rel( DI, ) Rel( D, ) D R( I) I where Rel(D, I) s the judged relevance, R(I) s the set of all relevant documents for topc I. (Ths year, for some of the nput judgments, {0, 1, 2} 3 levels of relevance s used.) Because judged rrelevant documents have relevance score 0, they wll not have any effect n the model. (1)
3 For the pseudo relevant documents, D D P, we have to use Bayes rule to expand the P(D I) term just as n (1), thus, arrvng at the followng decomposton, PI ( Dj) PD ( j) Pr ( I) = Pr ( D) PD ( I) + Pr ( D) PI ( ) D j Dj DP = Pr ( D) PD ( I) + Pr ( D) PI ( D) PD ( ) PI ( ) j j j DP Ths fnal score P(r I) wll be the weght for the expanson term r n the relevance model. The above dervatons are exact, except there are some unknowns n the result: P(I), P(D j ), and the total collectonrelevance Rel( D, ) D C I. Notce that these probabltes may depend on topc, and specal care must be taken to deal wth these values, as they drectly affect the relatve mportance of true relevant documents w.r.t. pseudo relevant documents. We propose a justfed and effectve way of dealng wth ths balancng. 2.1 Balancng weghts from pseudo- and true- relevant documents There are several key problems to solve n order to properly balance the weghts. Frstly, snce some of the probabltes wll never be known, we want to use a hyper-parameter to weght the two sources, and tune the weght on tranng data. Secondly, the numbers of (true and pseudo) feedback documents can be dfferent across topcs, thus, a larger D T set would probably lead to a bas toward terms from the judged set and thus the optmal hyper parameter would vary. That s why a normalzaton step needs to happen so that the numbers of feedback documents true v.s. pseudo wll not affect the relatve weghts of the terms from these two sources. Thrdly, note P(I D j ) the query generaton probablty wll have large varatons across topcs, as a query can contan dfferent number of query terms whch can be popular or rare. Thus, we need to reduce ths varance so that the hyper parameter s less varable across topcs, and t s easer to learn an effectve hyper parameter value whch wll be appled across all topcs. From equaton (3), snce we don t know the judgments of all the documents, we estmate P(D I) on the gven samples D T, ths gves us the resultng term P ˆ( D ) Rel(, ) Rel(, ) I = D I D D D I, call t the T emprcal relevance dstrbuton. (4) Pr ( I) = Pr ( D) PD ˆ( I) + Pr ( D) PI ( D) PD ( ) PI ( ) ˆ j j j DP (3) We denote the hyper parameterα, and parameterze P(r I) n α as follows, Rel(, ) PI ( D) α D I j P ( r I) = αp( r D) + (1 α) P( r D ) P( D ) Rel(, ) P( I) D I DP j j (5) P(D j ) s the document pror, f we assume unform, t doesn t matter what exact value t takes, as tunng α wll effectvely cancel ths constant term out. For an analogy wth the emprcal relevance dstrbuton, we also use the emprcal pror, 1/ D P. These two emprcal dstrbutons act as normalzers to the two sources, ensurng the weghts are normalzed properly wth respect to the numbers of samples n both sources. The above solves the frst and second problems outlned at the begnnng of the subsecton. For the last problem P(I), we don t know how exactly should t be determned, but we do know t should be dependent on the collecton and on the topc, and also we expect t to provde proper balancng effect for the term P(I D j ), whch vares a lot across topcs. Gven the requrements, we can have several alternatve assgnments for P(I), a) max D C{P(I D)}, b) avg Dj TOPN {P(I D j )}. Intutvely, f we neglect α,
4 assgnment (a) means we assume that a relevant document s as good as the best scorng document n the collecton, whle assgnment (b) means t s as good as the average score of the top N results. We avoded usng the smplest P(I C) the query generaton probablty of the collecton model as an approxmaton for P(I), because P(I D j ) would vary more wldly than P(I C) dependng on how well the documents n the collecton are matchng wth the topcs, whle P(I C) s gnorant of the documents. PI ( ) = max PI ( D) (6) D C PI ( ) = avg PI ( D) (7) D TOPN Wthout experments, t s dffcult to see how the two dfferent assgnments would behave because there s a regulatng parameter α outsde of the scope of the assgnments that wll determne how much weght to put on the two sources (true and pseudo relevance feedback). In general, we should select assgnments that produce the most stable optmal regulatng α across all topcs, or more drectly select the best assgnment accordng to the fnal retreval effectveness measure (such as overall MAP etc.). 3 Experments 3.1 Tranng and test datasets We tran all the parameters n the model on the tranng set by parameter sweeps. To create the tranng set, topcs and assessments from the followng track were used: TREC Terabyte 2004, 2005 and 2006 tracks and Mllon Query track We requre each tranng topc to contan at least 40 judgments, and use 2/3 of the judgements as feedback documents and the rest 1/3 as evaluaton data. For the feedback judgments, at least 4 should be relevant and 5 non relevant. Other topcs are fltered out. We keep a rato of 3:1 for MQ topcs and Terabyte topcs just to mmc the testng condtons and evaluate on the whole set usng MAP. Total number of topcs s 296, wth 74 Terabyte topcs. Although for tranng effcency reasons we sampled a smaller subset of 113 tranng topcs from ths whole set. The selecton of the feedback documents s crucal for creatng a proper tranng set. However, snce we don t have the same results data from prevous tracks as the relevance feedback track organzers do, we can only try to be as close to the test data as possble. Snce the TREC judgment pool bases toward top ranked documents, although maybe not as hgh a bas as the feedback set for the test data, we stll smply used a 2/3 random sample of the judgments from the pool as feedback and the rest as evaluaton data. We fear that drectly usng top ranked results from the baselne run would bas the results too much, but we dd not try, wth the offcal test data now avalable we could try and report rankngs of the feedback documents for the tranng sets that we may use, and compare to that of the test set. 3.2 Parameter tunng and performance analyss Submtted results are shown n Table 1. All parameters were tuned on tranng set. Because of the large number of parameters to tune (especally condtoned on the dfferent setups), we fx all the rest parameters and tune only one parameter at a tme. We fx smoothng parameters frst for the baselne Set A algorthm, and then keep the same parameters for feedback runs. When tunng feedback runs, we only vary the number of feedback terms and feedback documents, α and the weght for the orgnal query. From the results, we can see that the optmal α s qute stable across dfferent feedback settngs, meanng we have done proper balancng. Also, we dd a more detaled per topc optmal α analyss on Set C and Set D, and results show that although the optmal α vares across topcs, there are 1/3 of the topcs where α can be 0-1 wthout changng performance; most of these topcs are ether too hard (MAP<0.05) or too easy (MAP>0.5). In 3/4 of the other topcs where α makes a dfference, optmal α > 0.5, meanng true relevance feedback s emphaszed more than the top rankng result. If we take the optmal MAP value for each topc, we get the upper bound for the performance we can get f we vary α, the upper bound on the tranng set turns out to be MAP = for α n the set {0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0} and wth all
5 Table 1. Submted results wth common parameters: dependency model ntal queres, phrase smoothng Drchlet wth mu=4000, and term smoothng mu=1700, #feedback terms = 50, orgnal query weght=0.7, max assgnment for P(I). Other parameters shown below: Runs: Runtag Stop-words #fb Docs other parameters set as runs CMURF08.C1 and D1, especally, the orgnal query weght s set as 0.7, whch restrcts the effects of the expanson query. When only wth one (the top) relevant document for feedback, on the test set, our model (α = 0.7) performs sgnfcantly better than relevance feedback alone ( α = 1), MAP v.s , wth sgnfcance level p < by a pared sgn test. Although t s not sgnfcantly better than pseudo relevance feedback alone (α = 0, MAP= ). We also show per topc how our results compare to the medan system n Fgure 1. The results are worse than the medan on average. And a comparson wth other systems show that the baselne s much worse than other systems. However, comparng the relatve gans n MAP by usng more relevance data (Table 1), the constructed tranng set and the offcal test data have a smlar trend, and the most performance gan s n the 10% range. On the tranng data, ths gan s acheved rght at set B, whle on the test data, the best performance s at set E. Ths suggests that a randomly sampled feedback document (what s done on the tranng topcs) s more nformatve than a top ranked relevant document (for the test case). Also, ths relatve gan s smaller than the best systems, and t mght be because of the worse baselne. More experments need to be done to re-establsh a baselne and confrm. 4 Conclusons α Tranng: MAP, RR, P@10 Offcal Test*: MAP RR P@10, mtc statap CMURF08.A1 Included 10 < , , , , , CMURF08.B1 Excluded , , , , , CMURF08.C1 Excluded , , , , , CMURF08.D1 Excluded , , , , , CMURF08.E1 Excluded , , , , , CMURF08.B2 Included , , , , , CMURF08.C2 Excluded , , , , , CMURF08.D2 Excluded , , , , , * Reported test metrcs MAP RR and P@10 are evaluated on the 31 Terabyte track topcs, the mtc and statap measures are evaluated on MQ topcs, wth 237 and 208 vald topcs respectvely. We presented a coherent model of ncorporatng pseudo relevance feedback together wth true relevance feedback. We showed that the dffculty les manly n balancng term weghts comng from the two dfferent sources. Several steps have been taken to deal wth weght balancng, and reasonable alternatve solutons exst. Results show that mergng relevance feedback and pseudo relevance feedback s sgnfcantly better than relevance feedback alone when feedback documents are few. Randomly sampled relevant document s carryng more nformaton than a top ranked relevant document, n terms of feedback performance.
6 Per Topc AP Dfference from Medan System Fgure 1. Comparng CMURF08.E1 wth the medan system. For future work, we would lke to explore ways of modelng negatve feedback, more fnely controlled term extracton, such as restrctng a wndow around each query term, to get more precse feedback terms. We are also nterested n nvestgatng how perturbng the feedback documents would affect results. References Donald Metzler, Trevor Strohman, Yun Zhou and W.B. Croft, Indr at TREC 2005: Terabyte Track. In Proceedngs of the 14th Text REtreval Conference (TREC). Jngjng L and Hongfe Yan, Pekng Unversty at the TREC 2006 Terabyte Track. In Proceedngs of the 15th Text REtreval Conference (TREC). J. J. Rocho, Relevance Feedback n Informaton Retreval. In The SMART Retreval System Experments n Automatc Document Processng, pages Xuanhu Wang, Hu Fang and Chengxang Zha, A Study of Methods for Negatve Relevance Feedback. In Proceedngs of the 31st annual nternatonal ACM SIGIR conference on Research and development n nformaton retreval (SIGIR '08), pages Vctor Lavrenko and W.B. Croft, Relevance-Based Language Models. Proceedngs of the 24th Annual nternatonal ACM SIGIR Conference on Research and Development n nformaton Retreval (SIGIR '01), pages Yangbo Zhu, Le Zhao and Jame Callan, Structured Queres for Legal Search. In Proceedngs of the 2007 Text REtreval Conference (TREC 2007).
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 informationHomework 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 informationCS47300: Web Information Search and Management
CS47300: Web Informaton Search and Management Probablstc Retreval Models Prof. Chrs Clfton 7 September 2018 Materal adapted from course created by Dr. Luo S, now leadng Albaba research group 14 Why probabltes
More informationSimulated Power of the Discrete Cramér-von Mises Goodness-of-Fit Tests
Smulated of the Cramér-von Mses Goodness-of-Ft Tests Steele, M., Chaselng, J. and 3 Hurst, C. School of Mathematcal and Physcal Scences, James Cook Unversty, Australan School of Envronmental Studes, Grffth
More informationLecture 4. Instructor: Haipeng Luo
Lecture 4 Instructor: Hapeng Luo In the followng lectures, we focus on the expert problem and study more adaptve algorthms. Although Hedge s proven to be worst-case optmal, one may wonder how well t would
More informationDepartment of Statistics University of Toronto STA305H1S / 1004 HS Design and Analysis of Experiments Term Test - Winter Solution
Department of Statstcs Unversty of Toronto STA35HS / HS Desgn and Analyss of Experments Term Test - Wnter - Soluton February, Last Name: Frst Name: Student Number: Instructons: Tme: hours. Ads: a non-programmable
More informationNUMERICAL DIFFERENTIATION
NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the
More informationA Robust Method for Calculating the Correlation Coefficient
A Robust Method for Calculatng the Correlaton Coeffcent E.B. Nven and C. V. Deutsch Relatonshps between prmary and secondary data are frequently quantfed usng the correlaton coeffcent; however, the tradtonal
More informationProbabilistic Information Retrieval CE-324: Modern Information Retrieval Sharif University of Technology
Probablstc Informaton Retreval CE-324: Modern Informaton Retreval Sharf Unversty of Technology M. Soleyman Fall 2016 Most sldes have been adapted from: Profs. Mannng, Nayak & Raghavan (CS-276, Stanford)
More informationEvaluation for sets of classes
Evaluaton for Tet Categorzaton Classfcaton accuracy: usual n ML, the proporton of correct decsons, Not approprate f the populaton rate of the class s low Precson, Recall and F 1 Better measures 21 Evaluaton
More informationProblem 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 informationLecture 10: May 6, 2013
TTIC/CMSC 31150 Mathematcal Toolkt Sprng 013 Madhur Tulsan Lecture 10: May 6, 013 Scrbe: Wenje Luo In today s lecture, we manly talked about random walk on graphs and ntroduce the concept of graph expander,
More informationTopic 23 - Randomized Complete Block Designs (RCBD)
Topc 3 ANOVA (III) 3-1 Topc 3 - Randomzed Complete Block Desgns (RCBD) Defn: A Randomzed Complete Block Desgn s a varant of the completely randomzed desgn (CRD) that we recently learned. In ths desgn,
More informationGlobal Sensitivity. Tuesday 20 th February, 2018
Global Senstvty Tuesday 2 th February, 28 ) Local Senstvty Most senstvty analyses [] are based on local estmates of senstvty, typcally by expandng the response n a Taylor seres about some specfc values
More informationLecture 12: Discrete Laplacian
Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly
More informationGeneralized 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 informationRetrieval Models: Language models
CS-590I Informaton Retreval Retreval Models: Language models Luo S Department of Computer Scence Purdue Unversty Introducton to language model Ungram language model Document language model estmaton Maxmum
More informationSemi-supervised Classification with Active Query Selection
Sem-supervsed Classfcaton wth Actve Query Selecton Jao Wang and Swe Luo School of Computer and Informaton Technology, Beng Jaotong Unversty, Beng 00044, Chna Wangjao088@63.com Abstract. Labeled samples
More informationBayesian Learning. Smart Home Health Analytics Spring Nirmalya Roy Department of Information Systems University of Maryland Baltimore County
Smart Home Health Analytcs Sprng 2018 Bayesan Learnng Nrmalya Roy Department of Informaton Systems Unversty of Maryland Baltmore ounty www.umbc.edu Bayesan Learnng ombnes pror knowledge wth evdence to
More informationSelecting Good Expansion Terms for Pseudo-Relevance Feedback
Guhong Cao, Jan-Yun Ne Department of Computer Scence and Operatons Research Unversty of Montreal, Canada {caogu, ne}@ro.umontreal.ca Selectng Good Expanson erms for Pseudo-Relevance Feedback ABSRAC Pseudo-relevance
More informationChapter 13: Multiple Regression
Chapter 13: Multple Regresson 13.1 Developng the multple-regresson Model The general model can be descrbed as: It smplfes for two ndependent varables: The sample ft parameter b 0, b 1, and b are used to
More informationStanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011
Stanford Unversty CS359G: Graph Parttonng and Expanders Handout 4 Luca Trevsan January 3, 0 Lecture 4 In whch we prove the dffcult drecton of Cheeger s nequalty. As n the past lectures, consder an undrected
More informationSpeeding 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 informationCollege 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 informationCHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE
CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE Analytcal soluton s usually not possble when exctaton vares arbtrarly wth tme or f the system s nonlnear. Such problems can be solved by numercal tmesteppng
More informationNegative Binomial Regression
STATGRAPHICS Rev. 9/16/2013 Negatve Bnomal Regresson Summary... 1 Data Input... 3 Statstcal Model... 3 Analyss Summary... 4 Analyss Optons... 7 Plot of Ftted Model... 8 Observed Versus Predcted... 10 Predctons...
More informationLecture 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 information1 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 informationprinceton univ. F 13 cos 521: Advanced Algorithm Design Lecture 3: Large deviations bounds and applications Lecturer: Sanjeev Arora
prnceton unv. F 13 cos 521: Advanced Algorthm Desgn Lecture 3: Large devatons bounds and applcatons Lecturer: Sanjeev Arora Scrbe: Today s topc s devaton bounds: what s the probablty that a random varable
More informationParametric fractional imputation for missing data analysis. Jae Kwang Kim Survey Working Group Seminar March 29, 2010
Parametrc fractonal mputaton for mssng data analyss Jae Kwang Km Survey Workng Group Semnar March 29, 2010 1 Outlne Introducton Proposed method Fractonal mputaton Approxmaton Varance estmaton Multple mputaton
More informationQuestion Classification Using Language Modeling
Queston Classfcaton Usng Language Modelng We L Center for Intellgent Informaton Retreval Department of Computer Scence Unversty of Massachusetts, Amherst, MA 01003 ABSTRACT Queston classfcaton assgns a
More informationUnified Subspace Analysis for Face Recognition
Unfed Subspace Analyss for Face Recognton Xaogang Wang and Xaoou Tang Department of Informaton Engneerng The Chnese Unversty of Hong Kong Shatn, Hong Kong {xgwang, xtang}@e.cuhk.edu.hk Abstract PCA, LDA
More informationChapter 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 informationComputing MLE Bias Empirically
Computng MLE Bas Emprcally Kar Wa Lm Australan atonal Unversty January 3, 27 Abstract Ths note studes the bas arses from the MLE estmate of the rate parameter and the mean parameter of an exponental dstrbuton.
More information} Often, when learning, we deal with uncertainty:
Uncertanty and Learnng } Often, when learnng, we deal wth uncertanty: } Incomplete data sets, wth mssng nformaton } Nosy data sets, wth unrelable nformaton } Stochastcty: causes and effects related non-determnstcally
More information/ n ) are compared. The logic is: if the two
STAT C141, Sprng 2005 Lecture 13 Two sample tests One sample tests: examples of goodness of ft tests, where we are testng whether our data supports predctons. Two sample tests: called as tests of ndependence
More informationBoostrapaggregating (Bagging)
Boostrapaggregatng (Baggng) An ensemble meta-algorthm desgned to mprove the stablty and accuracy of machne learnng algorthms Can be used n both regresson and classfcaton Reduces varance and helps to avod
More informationEcon107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)
I. Classcal Assumptons Econ7 Appled Econometrcs Topc 3: Classcal Model (Studenmund, Chapter 4) We have defned OLS and studed some algebrac propertes of OLS. In ths topc we wll study statstcal propertes
More informationLOW 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 informationFeature 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 information3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X
Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number
More informationA Note on Bound for Jensen-Shannon Divergence by Jeffreys
OPEN ACCESS Conference Proceedngs Paper Entropy www.scforum.net/conference/ecea- A Note on Bound for Jensen-Shannon Dvergence by Jeffreys Takuya Yamano, * Department of Mathematcs and Physcs, Faculty of
More informationPop-Click Noise Detection Using Inter-Frame Correlation for Improved Portable Auditory Sensing
Advanced Scence and Technology Letters, pp.164-168 http://dx.do.org/10.14257/astl.2013 Pop-Clc Nose Detecton Usng Inter-Frame Correlaton for Improved Portable Audtory Sensng Dong Yun Lee, Kwang Myung Jeon,
More informationModule 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 informationNumerical Heat and Mass Transfer
Master degree n Mechancal Engneerng Numercal Heat and Mass Transfer 06-Fnte-Dfference Method (One-dmensonal, steady state heat conducton) Fausto Arpno f.arpno@uncas.t Introducton Why we use models and
More informationCS : Algorithms and Uncertainty Lecture 17 Date: October 26, 2016
CS 29-128: Algorthms and Uncertanty Lecture 17 Date: October 26, 2016 Instructor: Nkhl Bansal Scrbe: Mchael Denns 1 Introducton In ths lecture we wll be lookng nto the secretary problem, and an nterestng
More information4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA
4 Analyss of Varance (ANOVA) 5 ANOVA 51 Introducton ANOVA ANOVA s a way to estmate and test the means of multple populatons We wll start wth one-way ANOVA If the populatons ncluded n the study are selected
More informationOutline. Communication. Bellman Ford Algorithm. Bellman Ford Example. Bellman Ford Shortest Path [1]
DYNAMIC SHORTEST PATH SEARCH AND SYNCHRONIZED TASK SWITCHING Jay Wagenpfel, Adran Trachte 2 Outlne Shortest Communcaton Path Searchng Bellmann Ford algorthm Algorthm for dynamc case Modfcatons to our algorthm
More informationCanonical transformations
Canoncal transformatons November 23, 2014 Recall that we have defned a symplectc transformaton to be any lnear transformaton M A B leavng the symplectc form nvarant, Ω AB M A CM B DΩ CD Coordnate transformatons,
More informationPredictive Analytics : QM901.1x Prof U Dinesh Kumar, IIMB. All Rights Reserved, Indian Institute of Management Bangalore
Sesson Outlne Introducton to classfcaton problems and dscrete choce models. Introducton to Logstcs Regresson. Logstc functon and Logt functon. Maxmum Lkelhood Estmator (MLE) for estmaton of LR parameters.
More informationStatistics II Final Exam 26/6/18
Statstcs II Fnal Exam 26/6/18 Academc Year 2017/18 Solutons Exam duraton: 2 h 30 mn 1. (3 ponts) A town hall s conductng a study to determne the amount of leftover food produced by the restaurants n the
More informationThe 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 informationChapter 5 Multilevel Models
Chapter 5 Multlevel Models 5.1 Cross-sectonal multlevel models 5.1.1 Two-level models 5.1.2 Multple level models 5.1.3 Multple level modelng n other felds 5.2 Longtudnal multlevel models 5.2.1 Two-level
More informationComparison of Regression Lines
STATGRAPHICS Rev. 9/13/2013 Comparson of Regresson Lnes Summary... 1 Data Input... 3 Analyss Summary... 4 Plot of Ftted Model... 6 Condtonal Sums of Squares... 6 Analyss Optons... 7 Forecasts... 8 Confdence
More informationOne-sided finite-difference approximations suitable for use with Richardson extrapolation
Journal of Computatonal Physcs 219 (2006) 13 20 Short note One-sded fnte-dfference approxmatons sutable for use wth Rchardson extrapolaton Kumar Rahul, S.N. Bhattacharyya * Department of Mechancal Engneerng,
More informationBOOTSTRAP METHOD FOR TESTING OF EQUALITY OF SEVERAL MEANS. M. Krishna Reddy, B. Naveen Kumar and Y. Ramu
BOOTSTRAP METHOD FOR TESTING OF EQUALITY OF SEVERAL MEANS M. Krshna Reddy, B. Naveen Kumar and Y. Ramu Department of Statstcs, Osmana Unversty, Hyderabad -500 007, Inda. nanbyrozu@gmal.com, ramu0@gmal.com
More informationDifference 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 informationPulse Coded Modulation
Pulse Coded Modulaton PCM (Pulse Coded Modulaton) s a voce codng technque defned by the ITU-T G.711 standard and t s used n dgtal telephony to encode the voce sgnal. The frst step n the analog to dgtal
More information10.34 Fall 2015 Metropolis Monte Carlo Algorithm
10.34 Fall 2015 Metropols Monte Carlo Algorthm The Metropols Monte Carlo method s very useful for calculatng manydmensonal ntegraton. For e.g. n statstcal mechancs n order to calculate the prospertes of
More informationCSci 6974 and ECSE 6966 Math. Tech. for Vision, Graphics and Robotics Lecture 21, April 17, 2006 Estimating A Plane Homography
CSc 6974 and ECSE 6966 Math. Tech. for Vson, Graphcs and Robotcs Lecture 21, Aprl 17, 2006 Estmatng A Plane Homography Overvew We contnue wth a dscusson of the major ssues, usng estmaton of plane projectve
More informationUncertainty as the Overlap of Alternate Conditional Distributions
Uncertanty as the Overlap of Alternate Condtonal Dstrbutons Olena Babak and Clayton V. Deutsch Centre for Computatonal Geostatstcs Department of Cvl & Envronmental Engneerng Unversty of Alberta An mportant
More informationRegularized Discriminant Analysis for Face Recognition
1 Regularzed Dscrmnant Analyss for Face Recognton Itz Pma, Mayer Aladem Department of Electrcal and Computer Engneerng, Ben-Guron Unversty of the Negev P.O.Box 653, Beer-Sheva, 845, Israel. Abstract Ths
More informationCOMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS
Avalable onlne at http://sck.org J. Math. Comput. Sc. 3 (3), No., 6-3 ISSN: 97-537 COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS
More informationLinear 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 information8.6 The Complex Number System
8.6 The Complex Number System Earler n the chapter, we mentoned that we cannot have a negatve under a square root, snce the square of any postve or negatve number s always postve. In ths secton we want
More informationEvaluating ADM on a Four-Level Relevance Scale Document Set from NTCIR. (DRAFT - Not for Quotation)
Workng Notes of NTCIR-4, Tokyo, 2-4 June 24 Evaluatng on a Four-Level Relevance Scale Document Set from NTCIR (DRAFT - Not for Quotaton) Vncenzo Della Mea 1, Luca D Gaspero 2, Stefano Mzzaro 1 1 Department
More informationMarkov Chain Monte Carlo (MCMC), Gibbs Sampling, Metropolis Algorithms, and Simulated Annealing Bioinformatics Course Supplement
Markov Chan Monte Carlo MCMC, Gbbs Samplng, Metropols Algorthms, and Smulated Annealng 2001 Bonformatcs Course Supplement SNU Bontellgence Lab http://bsnuackr/ Outlne! Markov Chan Monte Carlo MCMC! Metropols-Hastngs
More informationLINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity
LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 30 Multcollnearty Dr. Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur 2 Remedes for multcollnearty Varous technques have
More informationErrors for Linear Systems
Errors for Lnear Systems When we solve a lnear system Ax b we often do not know A and b exactly, but have only approxmatons  and ˆb avalable. Then the best thng we can do s to solve ˆx ˆb exactly whch
More informationBIO Lab 2: TWO-LEVEL NORMAL MODELS with school children popularity data
Lab : TWO-LEVEL NORMAL MODELS wth school chldren popularty data Purpose: Introduce basc two-level models for normally dstrbuted responses usng STATA. In partcular, we dscuss Random ntercept models wthout
More informationCopyright 2017 by Taylor Enterprises, Inc., All Rights Reserved. Adjusted Control Limits for U Charts. Dr. Wayne A. Taylor
Taylor Enterprses, Inc. Adjusted Control Lmts for U Charts Copyrght 207 by Taylor Enterprses, Inc., All Rghts Reserved. Adjusted Control Lmts for U Charts Dr. Wayne A. Taylor Abstract: U charts are used
More informationA Network Intrusion Detection Method Based on Improved K-means Algorithm
Advanced Scence and Technology Letters, pp.429-433 http://dx.do.org/10.14257/astl.2014.53.89 A Network Intruson Detecton Method Based on Improved K-means Algorthm Meng Gao 1,1, Nhong Wang 1, 1 Informaton
More informationComputation of Higher Order Moments from Two Multinomial Overdispersion Likelihood Models
Computaton of Hgher Order Moments from Two Multnomal Overdsperson Lkelhood Models BY J. T. NEWCOMER, N. K. NEERCHAL Department of Mathematcs and Statstcs, Unversty of Maryland, Baltmore County, Baltmore,
More informationOn the correction of the h-index for career length
1 On the correcton of the h-ndex for career length by L. Egghe Unverstet Hasselt (UHasselt), Campus Depenbeek, Agoralaan, B-3590 Depenbeek, Belgum 1 and Unverstet Antwerpen (UA), IBW, Stadscampus, Venusstraat
More informationNatural 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 informationLecture 4 Hypothesis Testing
Lecture 4 Hypothess Testng We may wsh to test pror hypotheses about the coeffcents we estmate. We can use the estmates to test whether the data rejects our hypothess. An example mght be that we wsh to
More informationJoint Statistical Meetings - Biopharmaceutical Section
Iteratve Ch-Square Test for Equvalence of Multple Treatment Groups Te-Hua Ng*, U.S. Food and Drug Admnstraton 1401 Rockvlle Pke, #200S, HFM-217, Rockvlle, MD 20852-1448 Key Words: Equvalence Testng; Actve
More informationTime-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 information10-701/ 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 informationTemperature. Chapter Heat Engine
Chapter 3 Temperature In prevous chapters of these notes we ntroduced the Prncple of Maxmum ntropy as a technque for estmatng probablty dstrbutons consstent wth constrants. In Chapter 9 we dscussed the
More informationA Hybrid Variational Iteration Method for Blasius Equation
Avalable at http://pvamu.edu/aam Appl. Appl. Math. ISSN: 1932-9466 Vol. 10, Issue 1 (June 2015), pp. 223-229 Applcatons and Appled Mathematcs: An Internatonal Journal (AAM) A Hybrd Varatonal Iteraton Method
More informationANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)
Econ 413 Exam 13 H ANSWERS Settet er nndelt 9 deloppgaver, A,B,C, som alle anbefales å telle lkt for å gøre det ltt lettere å stå. Svar er gtt . Unfortunately, there s a prntng error n the hnt of
More information1 GSW Iterative Techniques for y = Ax
1 for y = A I m gong to cheat here. here are a lot of teratve technques that can be used to solve the general case of a set of smultaneous equatons (wrtten n the matr form as y = A), but ths chapter sn
More informationVQ widely used in coding speech, image, and video
at Scalar quantzers are specal cases of vector quantzers (VQ): they are constraned to look at one sample at a tme (memoryless) VQ does not have such constrant better RD perfomance expected Source codng
More informationMDL-Based Unsupervised Attribute Ranking
MDL-Based Unsupervsed Attrbute Rankng Zdravko Markov Computer Scence Department Central Connectcut State Unversty New Brtan, CT 06050, USA http://www.cs.ccsu.edu/~markov/ markovz@ccsu.edu MDL-Based Unsupervsed
More informationStatistical Evaluation of WATFLOOD
tatstcal Evaluaton of WATFLD By: Angela MacLean, Dept. of Cvl & Envronmental Engneerng, Unversty of Waterloo, n. ctober, 005 The statstcs program assocated wth WATFLD uses spl.csv fle that s produced wth
More informationChapter 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 informationECE559VV Project Report
ECE559VV Project Report (Supplementary Notes Loc Xuan Bu I. MAX SUM-RATE SCHEDULING: THE UPLINK CASE We have seen (n the presentaton that, for downlnk (broadcast channels, the strategy maxmzng the sum-rate
More informationGaussian Mixture Models
Lab Gaussan Mxture Models Lab Objectve: Understand the formulaton of Gaussan Mxture Models (GMMs) and how to estmate GMM parameters. You ve already seen GMMs as the observaton dstrbuton n certan contnuous
More informationBayesian predictive Configural Frequency Analysis
Psychologcal Test and Assessment Modelng, Volume 54, 2012 (3), 285-292 Bayesan predctve Confgural Frequency Analyss Eduardo Gutérrez-Peña 1 Abstract Confgural Frequency Analyss s a method for cell-wse
More informationLecture 4: November 17, Part 1 Single Buffer Management
Lecturer: Ad Rosén Algorthms for the anagement of Networs Fall 2003-2004 Lecture 4: November 7, 2003 Scrbe: Guy Grebla Part Sngle Buffer anagement In the prevous lecture we taled about the Combned Input
More informationUncertainty in measurements of power and energy on power networks
Uncertanty n measurements of power and energy on power networks E. Manov, N. Kolev Department of Measurement and Instrumentaton, Techncal Unversty Sofa, bul. Klment Ohrdsk No8, bl., 000 Sofa, Bulgara Tel./fax:
More informationEngineering Risk Benefit Analysis
Engneerng Rsk Beneft Analyss.55, 2.943, 3.577, 6.938, 0.86, 3.62, 6.862, 22.82, ESD.72, ESD.72 RPRA 2. Elements of Probablty Theory George E. Apostolaks Massachusetts Insttute of Technology Sprng 2007
More informationLecture Randomized Load Balancing strategies and their analysis. Probability concepts include, counting, the union bound, and Chernoff bounds.
U.C. Berkeley CS273: Parallel and Dstrbuted Theory Lecture 1 Professor Satsh Rao August 26, 2010 Lecturer: Satsh Rao Last revsed September 2, 2010 Lecture 1 1 Course Outlne We wll cover a samplng of the
More informationIntroduction to Vapor/Liquid Equilibrium, part 2. Raoult s Law:
CE304, Sprng 2004 Lecture 4 Introducton to Vapor/Lqud Equlbrum, part 2 Raoult s Law: The smplest model that allows us do VLE calculatons s obtaned when we assume that the vapor phase s an deal gas, and
More informationSTAT 3008 Applied Regression Analysis
STAT 3008 Appled Regresson Analyss Tutoral : Smple Lnear Regresson LAI Chun He Department of Statstcs, The Chnese Unversty of Hong Kong 1 Model Assumpton To quantfy the relatonshp between two factors,
More informationDr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur
Analyss of Varance and Desgn of Experment-I MODULE VII LECTURE - 3 ANALYSIS OF COVARIANCE Dr Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur Any scentfc experment s performed
More informationNote on EM-training of IBM-model 1
Note on EM-tranng of IBM-model INF58 Language Technologcal Applcatons, Fall The sldes on ths subject (nf58 6.pdf) ncludng the example seem nsuffcent to gve a good grasp of what s gong on. Hence here are
More informationJAB Chain. Long-tail claims development. ASTIN - September 2005 B.Verdier A. Klinger
JAB Chan Long-tal clams development ASTIN - September 2005 B.Verder A. Klnger Outlne Chan Ladder : comments A frst soluton: Munch Chan Ladder JAB Chan Chan Ladder: Comments Black lne: average pad to ncurred
More informationSingular 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