COS 511: Theoretical Machine Learning. Lecturer: Rob Schapire Lecture #21 Scribe: Lawrence Diao April 23, 2013
|
|
- Duane Leslie Strickland
- 6 years ago
- Views:
Transcription
1 COS 511: Theoretcal Machne Learnng Lecturer: Rob Schapre Lecture #21 Scrbe: Lawrence Dao Aprl 23, On-Lne Log Loss To recap the end of the last lecture, we have the followng on-lne problem wth N experts. For each round t = 1,..., T : each expert predcts p t, dstrbuton on X master predcts q t dstrbuton on X observe x t X loss = ln q t (x t ) We want to get a bound on the total loss of the master q t n comparson to the best expert. log q t (x t ) mn log p t, (x t ) + small (1) where here we use the general log functon of arbtrary base. We ll see that ths on-lne log loss settng manfests tself n many applcatons such as horse racng and codng theory. 2 Codng Theory Here, we are concerned wth how to effcently send a message from Alce to Bob n as few bts as possble. In ths settng we defne X as the alphabet, and each x X as a letter. Say Alce wants to send one letter x. Defne p(x) to be the probablty of sendng x, whch you can estmate from a corpus. The best you can do s to take lg p(x) bts to send x. Now Alce s tryng to send a sequence of letters x 1, x 2, x 3,... One way we can do ths s to use p(x) for each letter separately, but ths s sub-optmal for Englsh. For example, f we see the followng strng of characters I am go, we can easly predct the next letter to be n gven the context, but f we smply use p(x), then we mght say that e s the most lkely, snce t s the letter of hghest frequency n the Englsh language. Our goal s to use the context to use fewer bts to encode x. If we defne p t (x t ) to be the probablty dstrbuton of x t gven context x t 1 1 = x 1,..., x t 1, then t takes lg p t (x t ) to encode the extra letter x t. However, t s really hard to model ths probablty. You can t get t just by countng as we could wth p(x). Instead, we consder combnng a collecton of codng methods where we don t know whch one wll be best. Let s say we have N codng methods (N experts). We try to pck a master codng method that uses at most a small amount more bts than the best encodng method.
2 Let p t, (x t ) = probablty of x t gven x t 1 1 accordng to the -th codng method. So we have lg p t, (x t ) bts used by -th codng method lg q t (x t ) bts used by arbtrary codng method q t We are tryng to come up wth a codng method q t (x t ) to guarantee lg q t (x t ) mn lg p t, (x t ) + small Such an algorthm s called a unversal compresson algorthm, snce t works about as well as the best codng method for any nput. Note that the bound should hold for any sequence of x t s, so there s no assumpton on randomness of x t. Also note that ths bound s of the form of (1). 3 Unversal Compresson Algorthm In ths secton we try to determne the algorthm for choosng the master codng method. To make the math cleaner, we change the base back to e, and try to acheve the followng bound ln q t (x t ) mn ln p t, (x t ) + small We also make the followng notaton changes q t (x t ) q(x t x t 1 p t, (x t ) p (x t x t 1 Let s pretend that x t are random even though they re not n order to motvate an algorthm for pckng q. Pretend that x t are pcked as follows: select one expert wth Pr[ = ] = 1 N x 1, x 2,..., generated accordng to : Pr[x 1 = ] = p (x Pr[x 2 x 1, = ] = p (x 2 x... 1, = ] = p (x t x t 1 2
3 Then the most natural way to pck q s: q(x t x t 1 = = Pr[x t, = x t 1 margnalze = = Pr[ = x t 1 Pr[x t =, x t 1 condtonal probablty w t, p [x t x t 1 w t, = Pr[ = x t 1 If we can fnd these w t, then we have an algorthm. w 1, = Pr[ = ] = 1 N w t+1, = Pr[ = x t 1] = Pr[ = x t 1 1, x t ] = Pr[ = x1 t Pr[x t =, x t 1 = w t, p (x t x1 t Normalzaton So we are left wth the followng algorthm. ntalzaton bayes rule : w 1, = 1 N On round t: Choose q(x t x t 1 = w t, p (x t x t 1 Update Weghts: : w t+1, = w t,p (x t x t 1 Normalzaton We can see that ths weght update s very smlar to other weght-update onlne learnng algorthms we have seen n the past, except we don t have to tune β snce there s only one correct choce of β = e 1 n ths case. 4 Boundng the Log Loss w t+1, w t, β loss loss = ln p t, (x t ) β = e 1 β loss = p t, (x t ) w t+1, w t, β loss = w t, p t, (x t ) Here we are tryng to prove (1), gven our choce of q(x t x t 1 = Theorem: log q t (x t ) mn log p t, (x t ) + log N 3
4 Defne q(x T = q(x q(x 2 x q(x 3 x 1, x 2 )... T = q(x t x t 1 = T = Pr[x T chan rule In the same way we can do ths wth each expert p (x T = Pr[x T = ] Addtonally, the total loss of our algorthm s gven by the followng: Smlarly, for any expert, log q t (x t ) = t [ = log = log q(x T log q(x t x t 1 t q(x t x t 1 log p t, (x t ) = log p (x T So we have the followng bound: q(x T = Pr[x T = Pr[ = ] Pr[x T = ] margnalze ] = 1 p (x T N 1 N p (x T = log q(x T log p (x T + log N = log q t (x t ) mn log p t, (x t ) + log N Here we consder log N to be small. Note that ths bound does not assume any randomness for x t. Now, let s consder an alternatve encodng scheme, where Alce wats for the entre message x 1, x 2,..., x T, chooses the best out of the N canddate encodng methods, uses lg N bts to encode whch encodng method she used, and fnally sends her message accordng to ths chosen method. We can see that ths scheme would use just as many bts as the rght hand sde of the bound, but usng our onlne algorthm we don t have to wat for the whole message to start encodng/sendng. We won t go nto detal about decodng, but n order to decode, Bob effectvely just smulates what Alce does to encode, so decodng s just as effcent as Alce s encodng, makng algorthmc effcency a non-factor. 4
5 5 Varatons 5.1 Usng a pror In ths secton we consder a pror Pr[ = ] = π not necessarly unform. Everythng about our algorthm stays the same except the ntal weghts are now w 1, = π, and the fnal bound ends up beng [ T ] log q t (x t ) mn log p t, (x t ) log π 5.2 Infnte Experts { 1 wth prob p Consder the problem where X = {0, 1}, and expert p predcts x t = 0 wth prob 1 p where we have all experts p [0, 1]. We need to fgure out the weghts w t,p to get q. In the fnte case, we had w t, = P r[ = x t 1, but applyng ths defnton to the nfnte case doesn t really make sense unless we re talkng about the probablty densty: Pr[p dp x t 1 = Pr[xt 1 1 p dp] P r[p dp] Pr[x t 1 = Pr[xt 1 1 p dp] Pr[p dp] Normalzaton = Pr[xt 1 1 p dp] Normalzaton p h (1 p) t h 1 bayes rule assumng Pr[p dp] unform where h s the number of heads (1 s) n the frst t 1 rounds. Now, lettng w t,p = p h (1 p) t h q t = w t,ppdp 1 0 w t,pdp Normalzaton = h + 1 (t 1) + 2 sometmes called laplace smoothng We can get a smlar bound as before n ths case but log π or lg N doesn t make sense. We ll see a bound n a future lecture. 6 Swtchng Experts In ths secton we set up the problem for next class. Here, we no longer assume that one expert s good all the tme. Instead, we change the model so that at any step, the correct expert can swtch to another expert. However, the learnng algorthm has no dea when the experts are swtchng. Our goal s to desgn an algorthm that performs well wth respect to the best swtchng sequence of experts. We ll look at ths n the next lecture. 5
Parallel Prefix addition
Marcelo Kryger Sudent ID 015629850 Parallel Prefx addton The parallel prefx adder presented next, performs the addton of two bnary numbers n tme of complexty O(log n) and lnear cost O(n). Lets notce the
More informationScribe: Chris Berlind Date: Feb 1, 2010
CS/CNS/EE 253: Advanced Topcs n Machne Learnng Topc: Dealng wth Partal Feedback #2 Lecturer: Danel Golovn Scrbe: Chrs Berlnd Date: Feb 1, 2010 8.1 Revew In the prevous lecture we began lookng at algorthms
More information15-451/651: Design & Analysis of Algorithms January 22, 2019 Lecture #3: Amortized Analysis last changed: January 18, 2019
5-45/65: Desgn & Analyss of Algorthms January, 09 Lecture #3: Amortzed Analyss last changed: January 8, 09 Introducton In ths lecture we dscuss a useful form of analyss, called amortzed analyss, for problems
More informationOCR Statistics 1 Working with data. Section 2: Measures of location
OCR Statstcs 1 Workng wth data Secton 2: Measures of locaton Notes and Examples These notes have sub-sectons on: The medan Estmatng the medan from grouped data The mean Estmatng the mean from grouped data
More informationFoundations of Machine Learning II TP1: Entropy
Foundatons of Machne Learnng II TP1: Entropy Gullaume Charpat (Teacher) & Gaétan Marceau Caron (Scrbe) Problem 1 (Gbbs nequalty). Let p and q two probablty measures over a fnte alphabet X. Prove that KL(p
More informationTCOM501 Networking: Theory & Fundamentals Final Examination Professor Yannis A. Korilis April 26, 2002
TO5 Networng: Theory & undamentals nal xamnaton Professor Yanns. orls prl, Problem [ ponts]: onsder a rng networ wth nodes,,,. In ths networ, a customer that completes servce at node exts the networ wth
More informationLecture 7. We now use Brouwer s fixed point theorem to prove Nash s theorem.
Topcs on the Border of Economcs and Computaton December 11, 2005 Lecturer: Noam Nsan Lecture 7 Scrbe: Yoram Bachrach 1 Nash s Theorem We begn by provng Nash s Theorem about the exstance of a mxed strategy
More informationTests for Two Ordered Categorical Variables
Chapter 253 Tests for Two Ordered Categorcal Varables Introducton Ths module computes power and sample sze for tests of ordered categorcal data such as Lkert scale data. Assumng proportonal odds, such
More informationApplications of Myerson s Lemma
Applcatons of Myerson s Lemma Professor Greenwald 28-2-7 We apply Myerson s lemma to solve the sngle-good aucton, and the generalzaton n whch there are k dentcal copes of the good. Our objectve s welfare
More informationTests for Two Correlations
PASS Sample Sze Software Chapter 805 Tests for Two Correlatons Introducton The correlaton coeffcent (or correlaton), ρ, s a popular parameter for descrbng the strength of the assocaton between two varables.
More informationAppendix - Normally Distributed Admissible Choices are Optimal
Appendx - Normally Dstrbuted Admssble Choces are Optmal James N. Bodurtha, Jr. McDonough School of Busness Georgetown Unversty and Q Shen Stafford Partners Aprl 994 latest revson September 00 Abstract
More informationFinancial mathematics
Fnancal mathematcs Jean-Luc Bouchot jean-luc.bouchot@drexel.edu February 19, 2013 Warnng Ths s a work n progress. I can not ensure t to be mstake free at the moment. It s also lackng some nformaton. But
More information2.1 Rademacher Calculus... 3
COS 598E: Unsupervsed Learnng Week 2 Lecturer: Elad Hazan Scrbe: Kran Vodrahall Contents 1 Introducton 1 2 Non-generatve pproach 1 2.1 Rademacher Calculus............................... 3 3 Spectral utoencoders
More informationElements of Economic Analysis II Lecture VI: Industry Supply
Elements of Economc Analyss II Lecture VI: Industry Supply Ka Hao Yang 10/12/2017 In the prevous lecture, we analyzed the frm s supply decson usng a set of smple graphcal analyses. In fact, the dscusson
More informationSurvey of Math: Chapter 22: Consumer Finance Borrowing Page 1
Survey of Math: Chapter 22: Consumer Fnance Borrowng Page 1 APR and EAR Borrowng s savng looked at from a dfferent perspectve. The dea of smple nterest and compound nterest stll apply. A new term s the
More informationRandom Variables. b 2.
Random Varables Generally the object of an nvestgators nterest s not necessarly the acton n the sample space but rather some functon of t. Techncally a real valued functon or mappng whose doman s the sample
More informationThe IBM Translation Models. Michael Collins, Columbia University
The IBM Translaton Models Mchael Collns, Columba Unversty Recap: The Nosy Channel Model Goal: translaton system from French to Englsh Have a model p(e f) whch estmates condtonal probablty of any Englsh
More informationCS 286r: Matching and Market Design Lecture 2 Combinatorial Markets, Walrasian Equilibrium, Tâtonnement
CS 286r: Matchng and Market Desgn Lecture 2 Combnatoral Markets, Walrasan Equlbrum, Tâtonnement Matchng and Money Recall: Last tme we descrbed the Hungaran Method for computng a maxmumweght bpartte matchng.
More informationA MODEL OF COMPETITION AMONG TELECOMMUNICATION SERVICE PROVIDERS BASED ON REPEATED GAME
A MODEL OF COMPETITION AMONG TELECOMMUNICATION SERVICE PROVIDERS BASED ON REPEATED GAME Vesna Radonć Đogatovć, Valentna Radočć Unversty of Belgrade Faculty of Transport and Traffc Engneerng Belgrade, Serba
More informationSupplementary material for Non-conjugate Variational Message Passing for Multinomial and Binary Regression
Supplementary materal for Non-conjugate Varatonal Message Passng for Multnomal and Bnary Regresson October 9, 011 1 Alternatve dervaton We wll focus on a partcular factor f a and varable x, wth the am
More informationMathematical Thinking Exam 1 09 October 2017
Mathematcal Thnkng Exam 1 09 October 2017 Name: Instructons: Be sure to read each problem s drectons. Wrte clearly durng the exam and fully erase or mark out anythng you do not want graded. You may use
More informationEconomics 1410 Fall Section 7 Notes 1. Define the tax in a flexible way using T (z), where z is the income reported by the agent.
Economcs 1410 Fall 2017 Harvard Unversty Yaan Al-Karableh Secton 7 Notes 1 I. The ncome taxaton problem Defne the tax n a flexble way usng T (), where s the ncome reported by the agent. Retenton functon:
More information- contrast so-called first-best outcome of Lindahl equilibrium with case of private provision through voluntary contributions of households
Prvate Provson - contrast so-called frst-best outcome of Lndahl equlbrum wth case of prvate provson through voluntary contrbutons of households - need to make an assumpton about how each household expects
More informationFinite Math - Fall Section Future Value of an Annuity; Sinking Funds
Fnte Math - Fall 2016 Lecture Notes - 9/19/2016 Secton 3.3 - Future Value of an Annuty; Snkng Funds Snkng Funds. We can turn the annutes pcture around and ask how much we would need to depost nto an account
More informationLikelihood Fits. Craig Blocker Brandeis August 23, 2004
Lkelhood Fts Crag Blocker Brandes August 23, 2004 Outlne I. What s the queston? II. Lkelhood Bascs III. Mathematcal Propertes IV. Uncertantes on Parameters V. Mscellaneous VI. Goodness of Ft VII. Comparson
More information/ Computational Genomics. Normalization
0-80 /02-70 Computatonal Genomcs Normalzaton Gene Expresson Analyss Model Computatonal nformaton fuson Bologcal regulatory networks Pattern Recognton Data Analyss clusterng, classfcaton normalzaton, mss.
More informationProblem Set 6 Finance 1,
Carnege Mellon Unversty Graduate School of Industral Admnstraton Chrs Telmer Wnter 2006 Problem Set 6 Fnance, 47-720. (representatve agent constructon) Consder the followng two-perod, two-agent economy.
More informationData Mining Linear and Logistic Regression
07/02/207 Data Mnng Lnear and Logstc Regresson Mchael L of 26 Regresson In statstcal modellng, regresson analyss s a statstcal process for estmatng the relatonshps among varables. Regresson models are
More informationTopics on the Border of Economics and Computation November 6, Lecture 2
Topcs on the Border of Economcs and Computaton November 6, 2005 Lecturer: Noam Nsan Lecture 2 Scrbe: Arel Procacca 1 Introducton Last week we dscussed the bascs of zero-sum games n strategc form. We characterzed
More informationAppendix for Solving Asset Pricing Models when the Price-Dividend Function is Analytic
Appendx for Solvng Asset Prcng Models when the Prce-Dvdend Functon s Analytc Ovdu L. Caln Yu Chen Thomas F. Cosmano and Alex A. Hmonas January 3, 5 Ths appendx provdes proofs of some results stated n our
More informationCh Rival Pure private goods (most retail goods) Non-Rival Impure public goods (internet service)
h 7 1 Publc Goods o Rval goods: a good s rval f ts consumpton by one person precludes ts consumpton by another o Excludable goods: a good s excludable f you can reasonably prevent a person from consumng
More informationFinance 402: Problem Set 1 Solutions
Fnance 402: Problem Set 1 Solutons Note: Where approprate, the fnal answer for each problem s gven n bold talcs for those not nterested n the dscusson of the soluton. 1. The annual coupon rate s 6%. A
More informationEquilibrium in Prediction Markets with Buyers and Sellers
Equlbrum n Predcton Markets wth Buyers and Sellers Shpra Agrawal Nmrod Megddo Benamn Armbruster Abstract Predcton markets wth buyers and sellers of contracts on multple outcomes are shown to have unque
More informationChapter 5 Student Lecture Notes 5-1
Chapter 5 Student Lecture Notes 5-1 Basc Busness Statstcs (9 th Edton) Chapter 5 Some Important Dscrete Probablty Dstrbutons 004 Prentce-Hall, Inc. Chap 5-1 Chapter Topcs The Probablty Dstrbuton of a Dscrete
More informationElton, Gruber, Brown, and Goetzmann. Modern Portfolio Theory and Investment Analysis, 7th Edition. Solutions to Text Problems: Chapter 9
Elton, Gruber, Brown, and Goetzmann Modern Portfolo Theory and Investment Analyss, 7th Edton Solutons to Text Problems: Chapter 9 Chapter 9: Problem In the table below, gven that the rskless rate equals
More informationHewlett Packard 10BII Calculator
Hewlett Packard 0BII Calculator Keystrokes for the HP 0BII are shown n the tet. However, takng a mnute to revew the Quk Start secton, below, wll be very helpful n gettng started wth your calculator. Note:
More informationHomework 9: due Monday, 27 October, 2008
PROBLEM ONE Homework 9: due Monday, 7 October, 008. (Exercses from the book, 6 th edton, 6.6, -3.) Determne the number of dstnct orderngs of the letters gven: (a) GUIDE (b) SCHOOL (c) SALESPERSONS. (Exercses
More informationOPERATIONS RESEARCH. Game Theory
OPERATIONS RESEARCH Chapter 2 Game Theory Prof. Bbhas C. Gr Department of Mathematcs Jadavpur Unversty Kolkata, Inda Emal: bcgr.umath@gmal.com 1.0 Introducton Game theory was developed for decson makng
More informationTaxation and Externalities. - Much recent discussion of policy towards externalities, e.g., global warming debate/kyoto
Taxaton and Externaltes - Much recent dscusson of polcy towards externaltes, e.g., global warmng debate/kyoto - Increasng share of tax revenue from envronmental taxaton 6 percent n OECD - Envronmental
More informationSingle-Item Auctions. CS 234r: Markets for Networks and Crowds Lecture 4 Auctions, Mechanisms, and Welfare Maximization
CS 234r: Markets for Networks and Crowds Lecture 4 Auctons, Mechansms, and Welfare Maxmzaton Sngle-Item Auctons Suppose we have one or more tems to sell and a pool of potental buyers. How should we decde
More informationProduction and Supply Chain Management Logistics. Paolo Detti Department of Information Engeneering and Mathematical Sciences University of Siena
Producton and Supply Chan Management Logstcs Paolo Dett Department of Informaton Engeneerng and Mathematcal Scences Unversty of Sena Convergence and complexty of the algorthm Convergence of the algorthm
More informationGames and Decisions. Part I: Basic Theorems. Contents. 1 Introduction. Jane Yuxin Wang. 1 Introduction 1. 2 Two-player Games 2
Games and Decsons Part I: Basc Theorems Jane Yuxn Wang Contents 1 Introducton 1 2 Two-player Games 2 2.1 Zero-sum Games................................ 3 2.1.1 Pure Strateges.............................
More informationRepeated Games against Budgeted Adversaries
Repeated Games aganst Budgeted Adversares Jacob Abernethy Dvson of Computer Scence UC Berkeley jake@cs.berkeley.edu Manfred K. Warmuth Department of Computer Scence UC Santa Cruz manfred@cse.ucsc.edu Abstract
More informationLinear Combinations of Random Variables and Sampling (100 points)
Economcs 30330: Statstcs for Economcs Problem Set 6 Unversty of Notre Dame Instructor: Julo Garín Sprng 2012 Lnear Combnatons of Random Varables and Samplng 100 ponts 1. Four-part problem. Go get some
More informationSIMPLE FIXED-POINT ITERATION
SIMPLE FIXED-POINT ITERATION The fed-pont teraton method s an open root fndng method. The method starts wth the equaton f ( The equaton s then rearranged so that one s one the left hand sde of the equaton
More informationII. Random Variables. Variable Types. Variables Map Outcomes to Numbers
II. Random Varables Random varables operate n much the same way as the outcomes or events n some arbtrary sample space the dstncton s that random varables are smply outcomes that are represented numercally.
More informationoccurrence of a larger storm than our culvert or bridge is barely capable of handling? (what is The main question is: What is the possibility of
Module 8: Probablty and Statstcal Methods n Water Resources Engneerng Bob Ptt Unversty of Alabama Tuscaloosa, AL Flow data are avalable from numerous USGS operated flow recordng statons. Data s usually
More information2) In the medium-run/long-run, a decrease in the budget deficit will produce:
4.02 Quz 2 Solutons Fall 2004 Multple-Choce Questons ) Consder the wage-settng and prce-settng equatons we studed n class. Suppose the markup, µ, equals 0.25, and F(u,z) = -u. What s the natural rate of
More information2.1 The Inverting Configuration
/3/0 secton _ The nertng confguraton /. The Inertng Confguraton eadng Assgnment: pp. 6876 One use of amps s to make amplfers! Ths seems rather obous, but remember an amp by tself has too much gan to be
More informationMutual Funds and Management Styles. Active Portfolio Management
utual Funds and anagement Styles ctve Portfolo anagement ctve Portfolo anagement What s actve portfolo management? How can we measure the contrbuton of actve portfolo management? We start out wth the CP
More informationRandom Variables. 8.1 What is a Random Variable? Announcements: Chapter 8
Announcements: Quz starts after class today, ends Monday Last chance to take probablty survey ends Sunday mornng. Next few lectures: Today, Sectons 8.1 to 8. Monday, Secton 7.7 and extra materal Wed, Secton
More informationProblems to be discussed at the 5 th seminar Suggested solutions
ECON4260 Behavoral Economcs Problems to be dscussed at the 5 th semnar Suggested solutons Problem 1 a) Consder an ultmatum game n whch the proposer gets, ntally, 100 NOK. Assume that both the proposer
More informationA Php 5,000 loan is being repaid in 10 yearly payments. If interest is 8% effective, find the annual payment. 1 ( ) 10) 0.
Amortzaton If a loan s repad on nstalment (whch s usually n equal amounts); then the loan s sad to be repad by the amortzaton method. Under ths method, each nstalment ncludes the repayment of prncpal and
More informationA Case Study for Optimal Dynamic Simulation Allocation in Ordinal Optimization 1
A Case Study for Optmal Dynamc Smulaton Allocaton n Ordnal Optmzaton Chun-Hung Chen, Dongha He, and Mchael Fu 4 Abstract Ordnal Optmzaton has emerged as an effcent technque for smulaton and optmzaton.
More informationUnderstanding Annuities. Some Algebraic Terminology.
Understandng Annutes Ma 162 Sprng 2010 Ma 162 Sprng 2010 March 22, 2010 Some Algebrac Termnology We recall some terms and calculatons from elementary algebra A fnte sequence of numbers s a functon of natural
More informationPrinciples of Finance
Prncples of Fnance Grzegorz Trojanowsk Lecture 6: Captal Asset Prcng Model Prncples of Fnance - Lecture 6 1 Lecture 6 materal Requred readng: Elton et al., Chapters 13, 14, and 15 Supplementary readng:
More informationMonte Carlo Rendering
Last Tme? Monte Carlo Renderng Monte-Carlo Integraton Probabltes and Varance Analyss of Monte-Carlo Integraton Monte-Carlo n Graphcs Stratfed Samplng Importance Samplng Advanced Monte-Carlo Renderng Monte-Carlo
More informationCHAPTER 3: BAYESIAN DECISION THEORY
CHATER 3: BAYESIAN DECISION THEORY Decson makng under uncertanty 3 rogrammng computers to make nference from data requres nterdscplnary knowledge from statstcs and computer scence Knowledge of statstcs
More informationCracking VAR with kernels
CUTTIG EDGE. PORTFOLIO RISK AALYSIS Crackng VAR wth kernels Value-at-rsk analyss has become a key measure of portfolo rsk n recent years, but how can we calculate the contrbuton of some portfolo component?
More informationCHAPTER 9 FUNCTIONAL FORMS OF REGRESSION MODELS
CHAPTER 9 FUNCTIONAL FORMS OF REGRESSION MODELS QUESTIONS 9.1. (a) In a log-log model the dependent and all explanatory varables are n the logarthmc form. (b) In the log-ln model the dependent varable
More informationMeasures of Spread IQR and Deviation. For exam X, calculate the mean, median and mode. For exam Y, calculate the mean, median and mode.
Part 4 Measures of Spread IQR and Devaton In Part we learned how the three measures of center offer dfferent ways of provdng us wth a sngle representatve value for a data set. However, consder the followng
More informationJeffrey Ely. October 7, This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License.
October 7, 2012 Ths work s lcensed under the Creatve Commons Attrbuton-NonCommercal-ShareAlke 3.0 Lcense. Recap We saw last tme that any standard of socal welfare s problematc n a precse sense. If we want
More informationEvaluating Performance
5 Chapter Evaluatng Performance In Ths Chapter Dollar-Weghted Rate of Return Tme-Weghted Rate of Return Income Rate of Return Prncpal Rate of Return Daly Returns MPT Statstcs 5- Measurng Rates of Return
More information3: Central Limit Theorem, Systematic Errors
3: Central Lmt Theorem, Systematc Errors 1 Errors 1.1 Central Lmt Theorem Ths theorem s of prme mportance when measurng physcal quanttes because usually the mperfectons n the measurements are due to several
More informationFast Laplacian Solvers by Sparsification
Spectral Graph Theory Lecture 19 Fast Laplacan Solvers by Sparsfcaton Danel A. Spelman November 9, 2015 Dsclamer These notes are not necessarly an accurate representaton of what happened n class. The notes
More informationTrading by estimating the forward distribution using quantization and volatility information
Tradng by estmatng the forward dstrbuton usng quantzaton and volatlty nformaton Attla Ceffer, Janos Levendovszky Abstract In ths paper, novel algorthms are developed for electronc tradng on fnancal tme
More informationStatic (or Simultaneous- Move) Games of Complete Information
Statc (or Smultaneous- Move) Games of Complete Informaton Nash Equlbrum Best Response Functon F. Valognes - Game Theory - Chp 3 Outlne of Statc Games of Complete Informaton Introducton to games Normal-form
More information332 Mathematical Induction Solutions for Chapter 14. for every positive integer n. Proof. We will prove this with mathematical induction.
33 Mathematcal Inducton. Solutons for Chapter. Prove that 3 n n n for every postve nteger n. Proof. We wll prove ths wth mathematcal nducton. Observe that f n, ths statement s, whch s obvously true. Consder
More informationQuiz on Deterministic part of course October 22, 2002
Engneerng ystems Analyss for Desgn Quz on Determnstc part of course October 22, 2002 Ths s a closed book exercse. You may use calculators Grade Tables There are 90 ponts possble for the regular test, or
More informationIntroduction to PGMs: Discrete Variables. Sargur Srihari
Introducton to : Dscrete Varables Sargur srhar@cedar.buffalo.edu Topcs. What are graphcal models (or ) 2. Use of Engneerng and AI 3. Drectonalty n graphs 4. Bayesan Networks 5. Generatve Models and Samplng
More informationThe Hiring Problem. Informationsteknologi. Institutionen för informationsteknologi
The Hrng Problem An agency gves you a lst of n persons You ntervew them one-by-one After each ntervew, you must mmedately decde f ths canddate should be hred You can change your mnd f a better one comes
More informationOption pricing and numéraires
Opton prcng and numérares Daro Trevsan Unverstà degl Stud d Psa San Mnato - 15 September 2016 Overvew 1 What s a numerare? 2 Arrow-Debreu model Change of numerare change of measure 3 Contnuous tme Self-fnancng
More informationIntroduction. Chapter 7 - An Introduction to Portfolio Management
Introducton In the next three chapters, we wll examne dfferent aspects of captal market theory, ncludng: Brngng rsk and return nto the pcture of nvestment management Markowtz optmzaton Modelng rsk and
More informationFORD MOTOR CREDIT COMPANY SUGGESTED ANSWERS. Richard M. Levich. New York University Stern School of Business. Revised, February 1999
FORD MOTOR CREDIT COMPANY SUGGESTED ANSWERS by Rchard M. Levch New York Unversty Stern School of Busness Revsed, February 1999 1 SETTING UP THE PROBLEM The bond s beng sold to Swss nvestors for a prce
More information4.4 Doob s inequalities
34 CHAPTER 4. MARTINGALES 4.4 Doob s nequaltes The frst nterestng consequences of the optonal stoppng theorems are Doob s nequaltes. If M n s a martngale, denote M n =max applen M. Theorem 4.8 If M n s
More informationLecture Note 2 Time Value of Money
Seg250 Management Prncples for Engneerng Managers Lecture ote 2 Tme Value of Money Department of Systems Engneerng and Engneerng Management The Chnese Unversty of Hong Kong Interest: The Cost of Money
More informationECE 586GT: Problem Set 2: Problems and Solutions Uniqueness of Nash equilibria, zero sum games, evolutionary dynamics
Unversty of Illnos Fall 08 ECE 586GT: Problem Set : Problems and Solutons Unqueness of Nash equlbra, zero sum games, evolutonary dynamcs Due: Tuesday, Sept. 5, at begnnng of class Readng: Course notes,
More informationOnline Appendix for Merger Review for Markets with Buyer Power
Onlne Appendx for Merger Revew for Markets wth Buyer Power Smon Loertscher Lesle M. Marx July 23, 2018 Introducton In ths appendx we extend the framework of Loertscher and Marx (forthcomng) to allow two
More informationarxiv:cond-mat/ v1 [cond-mat.other] 28 Nov 2004
arxv:cond-mat/0411699v1 [cond-mat.other] 28 Nov 2004 Estmatng Probabltes of Default for Low Default Portfolos Katja Pluto and Drk Tasche November 23, 2004 Abstract For credt rsk management purposes n general,
More informationMULTIPLE CURVE CONSTRUCTION
MULTIPLE CURVE CONSTRUCTION RICHARD WHITE 1. Introducton In the post-credt-crunch world, swaps are generally collateralzed under a ISDA Master Agreement Andersen and Pterbarg p266, wth collateral rates
More informationGeneral Examination in Microeconomic Theory. Fall You have FOUR hours. 2. Answer all questions
HARVARD UNIVERSITY DEPARTMENT OF ECONOMICS General Examnaton n Mcroeconomc Theory Fall 2010 1. You have FOUR hours. 2. Answer all questons PLEASE USE A SEPARATE BLUE BOOK FOR EACH QUESTION AND WRITE THE
More informationAn Application of Alternative Weighting Matrix Collapsing Approaches for Improving Sample Estimates
Secton on Survey Research Methods An Applcaton of Alternatve Weghtng Matrx Collapsng Approaches for Improvng Sample Estmates Lnda Tompkns 1, Jay J. Km 2 1 Centers for Dsease Control and Preventon, atonal
More informationA Bayesian Classifier for Uncertain Data
A Bayesan Classfer for Uncertan Data Bao Qn, Yun Xa Department of Computer Scence Indana Unversty - Purdue Unversty Indanapols, USA {baoqn, yxa}@cs.upu.edu Fang L Department of Mathematcal Scences Indana
More informationUniversal Multiparty Data Exchange and Secret Key Agreement
Unversal Multparty Data Exchange and Secret Key Agreement Hmanshu Tyag Shun Watanabe 1 arxv:1605.01033v2 [cs.it] 23 Jan 2017 Abstract Multple partes observng correlated data seek to recover each other
More informationAn asymmetry-similarity-measure-based neural fuzzy inference system
Fuzzy Sets and Systems 15 (005) 535 551 www.elsever.com/locate/fss An asymmetry-smlarty-measure-based neural fuzzy nference system Cheng-Jan Ln, Wen-Hao Ho Department of Computer Scence and Informaton
More informationarxiv: v2 [math.co] 6 Apr 2016
On the number of equvalence classes of nvertble Boolean functons under acton of permutaton of varables on doman and range arxv:1603.04386v2 [math.co] 6 Apr 2016 Marko Carć and Modrag Žvkovć Abstract. Let
More informationIntroduction to game theory
Introducton to game theory Lectures n game theory ECON5210, Sprng 2009, Part 1 17.12.2008 G.B. Ashem, ECON5210-1 1 Overvew over lectures 1. Introducton to game theory 2. Modelng nteractve knowledge; equlbrum
More informationOn the Complexity of Fair Coin Flipping
On the Complexty of Far Con Flppng ftach Hatner Nkolaos Makryanns Eran Omr Aprl 16, 2018 Abstract n ther breakthrough result, Moran et al. Journal of Cryptology 16 show how to construct an r-round two-party
More informationSimulation Budget Allocation for Further Enhancing the Efficiency of Ordinal Optimization
Dscrete Event Dynamc Systems: Theory and Applcatons, 10, 51 70, 000. c 000 Kluwer Academc Publshers, Boston. Manufactured n The Netherlands. Smulaton Budget Allocaton for Further Enhancng the Effcency
More informationPhysics 4A. Error Analysis or Experimental Uncertainty. Error
Physcs 4A Error Analyss or Expermental Uncertanty Slde Slde 2 Slde 3 Slde 4 Slde 5 Slde 6 Slde 7 Slde 8 Slde 9 Slde 0 Slde Slde 2 Slde 3 Slde 4 Slde 5 Slde 6 Slde 7 Slde 8 Slde 9 Slde 20 Slde 2 Error n
More informationEXAMINATIONS OF THE HONG KONG STATISTICAL SOCIETY
EXAMINATIONS OF THE HONG KONG STATISTICAL SOCIETY HIGHER CERTIFICATE IN STATISTICS, 2013 MODULE 7 : Tme seres and ndex numbers Tme allowed: One and a half hours Canddates should answer THREE questons.
More informationAC : THE DIAGRAMMATIC AND MATHEMATICAL APPROACH OF PROJECT TIME-COST TRADEOFFS
AC 2008-1635: THE DIAGRAMMATIC AND MATHEMATICAL APPROACH OF PROJECT TIME-COST TRADEOFFS Kun-jung Hsu, Leader Unversty Amercan Socety for Engneerng Educaton, 2008 Page 13.1217.1 Ttle of the Paper: The Dagrammatc
More informationQuantitative Portfolio Theory & Performance Analysis
550.447 Quanttatve ortfolo Theory & erformance Analyss Wee of March 4 & 11 (snow), 013 ast Algorthms, the Effcent ronter & the Sngle-Index Model Where we are Chapters 1-3 of AL: erformance, Rs and MT Chapters
More informationChapter 11: Optimal Portfolio Choice and the Capital Asset Pricing Model
Chapter 11: Optmal Portolo Choce and the CAPM-1 Chapter 11: Optmal Portolo Choce and the Captal Asset Prcng Model Goal: determne the relatonshp between rsk and return key to ths process: examne how nvestors
More informationA DUAL EXTERIOR POINT SIMPLEX TYPE ALGORITHM FOR THE MINIMUM COST NETWORK FLOW PROBLEM
Yugoslav Journal of Operatons Research Vol 19 (2009), Number 1, 157-170 DOI:10.2298/YUJOR0901157G A DUAL EXTERIOR POINT SIMPLEX TYPE ALGORITHM FOR THE MINIMUM COST NETWORK FLOW PROBLEM George GERANIS Konstantnos
More informationarxiv: v1 [cs.ds] 16 Jul 2015
The Complexty of All-swtches Strategy Improvement John Fearnley and Rahul Savan Unversty of Lverpool arxv:1507.04500v1 [cs.ds] 16 Jul 2015 Abstract. Strategy mprovement s a wdely-used and well-studed class
More informationThe evaluation method of HVAC system s operation performance based on exergy flow analysis and DEA method
The evaluaton method of HVAC system s operaton performance based on exergy flow analyss and DEA method Xng Fang, Xnqao Jn, Yonghua Zhu, Bo Fan Shangha Jao Tong Unversty, Chna Overvew 1. Introducton 2.
More informationAlternatives to Shewhart Charts
Alternatves to Shewhart Charts CUSUM & EWMA S Wongsa Overvew Revstng Shewhart Control Charts Cumulatve Sum (CUSUM) Control Chart Eponentally Weghted Movng Average (EWMA) Control Chart 2 Revstng Shewhart
More informationOptimal Black-Box Reductions Between Optimization Objectives
Optmal Black-Box Reductons Between Optmzaton Objectves Zeyuan Allen-Zhu zeyuan@csal.mt.edu Prnceton Unversty Elad Hazan ehazan@cs.prnceton.edu Prnceton Unversty arxv:163.56v3 [math.oc] May 16 frst crculated
More informationHow to Share a Secret, Infinitely
How to Share a Secret, Infntely Ilan Komargodsk Mon Naor Eylon Yogev Abstract Secret sharng schemes allow a dealer to dstrbute a secret pece of nformaton among several partes such that only qualfed subsets
More information