Lecture 8: September 26

Size: px
Start display at page:

Download "Lecture 8: September 26"

Transcription

1 10-704: Information Processing and Learning Fall 2016 Lectrer: Aarti Singh Lectre 8: September 26 Note: These notes are based on scribed notes from Spring15 offering of this corse. LaTeX template cortesy of UC Berkeley EECS dept. Disclaimer: These notes have not been sbjected to the sal scrtiny reserved for formal pblications. They may be distribted otside this class only with the permission of the Instrctor. 8.1 Review Maximm Entropy and Information Projection Last time we discssed that the problem of finding the maximm entropy distribtion constrained to lie in a sbset Q P is essentially eqivalent to finding the information projection of the niform distribtion onto Q, i.e. the distribtion in Q that is closest to niform in KL sense 1 max H(p) = min D(p ) p Q p Q If the set of constraints in Q are linear in p, i.e. of the form E p [f j (X)] eqal to or bonded by some constant, then the maximm entropy distribtion belongs to the exponential family: p (x) = exp( j jf j (x)) Z where the Lagrange parameters = { j } are chosen so that p meets the constraints. The information projection can be defined more generally with respect to any given base distribtion p 0 (x) (instead of niform): min D(p p 0) p Q If the set of constraints in Q are linear in p, i.e. of the form E p [f j (X)] eqal to or bonded by some constant, then the information projection distribtion belongs to the Gibbs family: p (x) = p 0 (x) exp( j jf j (x)) Z where the normalizing constant is the partition fnction: Z = x p 0 (x)e j jfj(x). 1 Here the niform distribtion is defined sch that all distribtions in Q are absoltely continos with respect to it. 8-1

2 8-2 Lectre 8: September Maximm Entropy Dality with Maximm Likelihood Estimation So far, we haven t talked abot data in the discssion of maximm entropy. Often the constraints on the distribtion are actally specified sing the data. For example, when we seek Maximm likelihood model in the exponential (Gibbs) family then we are essentially seeking the Maximm Entropy distribtion (Information Projection) given empirical constraints based on data. We will show this connection next. Consider the maximm likelihood model given data X 1,..., X n p ML(x) = argmax p (X i ) p 1 = argmin log p p (X i ) [ ] 1 = argmin Eˆp log p p (X) [ = argmin Eˆp log ˆp(X) p p (X) = argmin p D(ˆp p ) + H(ˆp) = argmin p D(ˆp p ), ] + Eˆp [ log ] 1 ˆp(X) since the soltion is eqivalent withot H(ˆp). Note that the final soltion is not the same as the projection. The following theorem relates maximm likelihood estimation in exponential family with base distribtion p 0 to information projection of p 0 onto a set of distribtions with constraints specified by the empirical mean of the sfficient statistics: Theorem 8.1 Dality Theorem Let α j = Eˆp [f j (X)], then p ML(x) = argmin p D(ˆp p ) = argmin p P E p[f j(x)]=α i D(p p 0 ) = pip (x) The theorem states that the distribtion belonging to the exponential family (with sfficient statistics f j (x) and base distribtion p 0 (x)) whose parameters maximize the likelihood of data, is the same as the information projection of p 0 (x) on to a set of distribtions with linear eqality constraints (specified by f j (x)) that are given by data. Proof: Since we know the information projection lies in the exponential family, all we need to show is that the s in the maximm likelihood model satisfy the empirical linear constraints. So lets analyze the s that achieve the maximm likelihood of the data. Recall that Z = x = argmax = argmax p 0 (x) exp[ j j f j (x)] p (X i ) = argmax [log p 0 (X i ) + j and log p (X i ) j f j (X i ) log Z ].

3 Lectre 8: September Taking derivative with respect to 1,, m, of the log likelihood fnction, we get that log p (X i ) = f j (X i ) n log Z j j = f j (X i ) n Z Z j = f j (X i ) n p 0 (x)f j (x) exp[ k f k (x)] Z x k = f j (X i ) n [ p 0(x) exp[ k kf k (x)] ]f j (x) Z x = p (x)f j (x) f j (X i ) n x At the maximizing ML the derivative is eqal to 0, so we get: = p (x)f ML j(x) = 1 f j (X i ) n x = E p ML [f j (X)] = Eˆp [f j (X)] 8.3 Maximm Entropy Generalization and Dality with reglarized Maximm Likelihood We can consider a generalization of the maximm entropy (information projection) problem [DPS08] min D(p p 0) + U(E p [f]), p P where U(E p [f]) is a reglarizer and f = [f 1 (X)... f m (X)]. Here are three example reglarizers: Example 8.2 Standard Maximm entropy/information projection is obtained with U(E p [f]) = 1(E p [f] = Eˆp [f]) Notice that for the eqivalence to hold the indicator fnction is defined so that 1 A is 0 if A is tre and otherwise. This penalty reqires the tre constraints to match the empirical constraints exactly. Example 8.3 L1 Norm Reglarizer U(E p [f]) = 1( E p [f j ] Eˆp [f j ] β j ) j Here also we se the same definition of indicator fnction as above. This penalty reqires the tre constraints to match the empirical constraints in an l 1 sense.

4 8-4 Lectre 8: September 26 Example 8.4 L2 Norm Reglarizer U(E p [f]) = E p[f] Eˆp [f] 2 2α This penalty reqires the tre constraints to match the empirical constraints in l 2 sense. To find the soltion to the generalized MaxEnt problem, we cold consider taking the derivative of the reglarized objective with respect to p, however notice that some of the reglarizations are not differentiable. So far, we have mostly ignored sch isses, assming differentiability. Bt lets consider a more formal treatment via Fenchel dality (instead of Lagrangian dality) that allows s to handle convex bt nondifferentiable fnctions. First, lets define the convex conjgate or Fenchel dal of a fnction ψ(p) as ψ () = sp[ p ψ(p)]. p It is essentially the largest difference between a line throgh the origin with slope and the graph of the fnction. If the fnction is differentiable, the largest difference happens at a point p where the gradient of the fnction ψ (p ) =. See the image below, for example. For a convex fnction, the conjgate is jst a characterization of the fnction in terms of (intercept vales of) its spporting hyperplanes corresponding to different slopes. The following theorem relates a primal optimization problem of closed, proper and convex fnction(s) to the dal optimization problem specified in terms of convex conjgate of the fnction(s). Recall that a fnction is proper if it is not infinite everywhere. Definition 8.5 Fenchel s Dality Let ψ, ϕ be closed, proper, and convex, and A is any matrix. Fenchel s Dality states that inf p ψ(p) + ϕ(ap) = sp ψ (A T ) ϕ ( ) Retrning to the previos maximm entropy generalization problem. We can consider p(x) p x as a vector, which may be an infinite-dimensional object. Lets define a matrix F with entries F jx = f j (x). Then, F p = x f j(x)p(x) = E[f j (X)] and we have the primal min D(p p 0) + U(F p) p P

5 Lectre 8: September where U will be closed, convex, and proper. Let ψ(p) = D(p p 0 ) if p P and otherwise, which is closed, proper and convex in p. To apply Fenchel dality, we first derive the conjgate of ψ(p) as ψ () = ln( x p 0 (x)e x ). For closed, convex and proper fnctions, the conjgate of a conjgate is the fnction itself, hence we instead evalate ψ (p) = sp[ p ln( x p 0 (x)e x )] Taking derivative with respect to x and setting it eqal to 0, we get that optimal x satisfies p x = p 0(x)e x. Plgging this vale of we get: x p0(x)ex ψ (p) = p ln( x p 0 (x)e x ) = x p x ( x ln( x p 0 (x)e x )) = x p x ln e x x p 0(x)e x = x p x ln p x p 0 (x) = D(p p 0) = ψ(p) So, sing Fenchel dality we have the dal problem sp[ ψ (F ) U ( )] = sp[ ln p 0 (x)e (F ) x ] U ( ) = sp[ ln p 0 (x)e j jfj(x) U ( ) = sp[ ln Z U ( )] We will show that this dal problem is essentially finding the reglarized Maximm Likelihood model nder exponential family with base distribtion p 0 (x). Before we can do that, we need one more notion - that of a shifted reglarizer Shifted reglarization Define a shifted reglarizer with respect to any distribtion t as follows Then the dal of the shifted reglarizer is U t () = U(E t [f] ), Ut () = sp[ U t ()] = sp[ U(E t [f] )] = sp [ E t [f] U( )] = E t [f] + U ( )

6 8-6 Lectre 8: September Dal as Reglarized Maximm Likelihood Let Q() = ln Z U ( ), then Q() = ln Z U ( ) = ln Z U t () + E t [f] = E t [ln p 0 ] + E t [ln p 0 + f ln Z ] U t () = E t [ln p 0 ] + E t [ln p 0 exp( j jf j (x) ] Ut () = L t (0) L t () U t (), where L t () := E t [ln p ] is the loss of the exponential family model p (x) = p0 exp( j jfj(x) Z with respect to distribtion t. If t = ˆp, then this it jst the negative log likelihood of the data nder the model p. Therefore, the dal problem is sp Q() min L t () + Ut (). and when t = ˆp, this is jst reglarized Maximm likelihood estimation. We then look at some examples of how the reglarization on maximm entropy/information projection transforms to reglarization term in maximm likelihood soltion. Examples: 1. U(E p [f]) = I(E p [f] = E p [f]). Then, Z U p (E p [f]) = 1(E p [f] = 0). (8.1) U p () = sp[ 1( = 0)] = 0. (8.2) The last step follow since 1( = 0) is infinity everywhere except when = 0. The problem ths gets back to the basic maximm entropy dality with nreglarized maximm likelihood. 2. U(E p [f]) = I( E p [f j ] E p [f j ] β j, j). Then, U p (E p [f]) = 1( E p [f j ] β j, j). (8.3) U p () = sp[ 1( j β j )] = j β j j, (8.4) The last step follow since 1( j β j ) is infinity everywhere except when j β j. Ths the expression is maximized when j = sign( j )β j. This corresponds to maximm likelihood with l 1 reglarization. 3. U(E p [f]) = ( E p [f] E p [f] 2 2/2α. Then, U p (E p [f]) = E p [f] 2 2/2α. (8.5) U p () = sp[ /2α] = α 2 2/2, (8.6) The last step follows since the maximizing = α (in this case, penalty is differentiable - simply take derivative wrt and set to zero). This corresponds to maximm likelihood with l 2 2 reglarization.

7 Lectre 8: September References [DPS08] M. Ddik, S.J. Phillips and R. Schapire, Maximm Entropy Density Estimation with Generalized Reglarization and an Application to Species Distribtion Modeling, Jornal of Machine Learning Research 8, 2007, pp

Lecture 13: Duality Uses and Correspondences

Lecture 13: Duality Uses and Correspondences 10-725/36-725: Conve Optimization Fall 2016 Lectre 13: Dality Uses and Correspondences Lectrer: Ryan Tibshirani Scribes: Yichong X, Yany Liang, Yanning Li Note: LaTeX template cortesy of UC Berkeley EECS

More information

Essentials of optimal control theory in ECON 4140

Essentials of optimal control theory in ECON 4140 Essentials of optimal control theory in ECON 4140 Things yo need to know (and a detail yo need not care abot). A few words abot dynamic optimization in general. Dynamic optimization can be thoght of as

More information

BIOSTATISTICAL METHODS

BIOSTATISTICAL METHODS BIOSTATISTICAL METHOS FOR TRANSLATIONAL & CLINICAL RESEARCH ROC Crve: IAGNOSTIC MEICINE iagnostic tests have been presented as alwas having dichotomos otcomes. In some cases, the reslt of the test ma be

More information

Admissibility under the LINEX loss function in non-regular case. Hidekazu Tanaka. Received November 5, 2009; revised September 2, 2010

Admissibility under the LINEX loss function in non-regular case. Hidekazu Tanaka. Received November 5, 2009; revised September 2, 2010 Scientiae Mathematicae Japonicae Online, e-2012, 427 434 427 Admissibility nder the LINEX loss fnction in non-reglar case Hidekaz Tanaka Received November 5, 2009; revised September 2, 2010 Abstract. In

More information

ON OPTIMALITY CONDITIONS FOR ABSTRACT CONVEX VECTOR OPTIMIZATION PROBLEMS

ON OPTIMALITY CONDITIONS FOR ABSTRACT CONVEX VECTOR OPTIMIZATION PROBLEMS J. Korean Math. Soc. 44 2007), No. 4, pp. 971 985 ON OPTIMALITY CONDITIONS FOR ABSTRACT CONVEX VECTOR OPTIMIZATION PROBLEMS Ge Myng Lee and Kwang Baik Lee Reprinted from the Jornal of the Korean Mathematical

More information

Formal Methods for Deriving Element Equations

Formal Methods for Deriving Element Equations Formal Methods for Deriving Element Eqations And the importance of Shape Fnctions Formal Methods In previos lectres we obtained a bar element s stiffness eqations sing the Direct Method to obtain eact

More information

FEA Solution Procedure

FEA Solution Procedure EA Soltion Procedre (demonstrated with a -D bar element problem) EA Procedre for Static Analysis. Prepare the E model a. discretize (mesh) the strctre b. prescribe loads c. prescribe spports. Perform calclations

More information

10-704: Information Processing and Learning Spring Lecture 8: Feb 5

10-704: Information Processing and Learning Spring Lecture 8: Feb 5 10-704: Information Processing and Learning Spring 2015 Lecture 8: Feb 5 Lecturer: Aarti Singh Scribe: Siheng Chen Disclaimer: These notes have not been subjected to the usual scrutiny reserved for formal

More information

Bayes and Naïve Bayes Classifiers CS434

Bayes and Naïve Bayes Classifiers CS434 Bayes and Naïve Bayes Classifiers CS434 In this lectre 1. Review some basic probability concepts 2. Introdce a sefl probabilistic rle - Bayes rle 3. Introdce the learning algorithm based on Bayes rle (ths

More information

CHARACTERIZATIONS OF EXPONENTIAL DISTRIBUTION VIA CONDITIONAL EXPECTATIONS OF RECORD VALUES. George P. Yanev

CHARACTERIZATIONS OF EXPONENTIAL DISTRIBUTION VIA CONDITIONAL EXPECTATIONS OF RECORD VALUES. George P. Yanev Pliska Std. Math. Blgar. 2 (211), 233 242 STUDIA MATHEMATICA BULGARICA CHARACTERIZATIONS OF EXPONENTIAL DISTRIBUTION VIA CONDITIONAL EXPECTATIONS OF RECORD VALUES George P. Yanev We prove that the exponential

More information

Constrained Optimization and Lagrangian Duality

Constrained Optimization and Lagrangian Duality CIS 520: Machine Learning Oct 02, 2017 Constrained Optimization and Lagrangian Duality Lecturer: Shivani Agarwal Disclaimer: These notes are designed to be a supplement to the lecture. They may or may

More information

The Linear Quadratic Regulator

The Linear Quadratic Regulator 10 The Linear Qadratic Reglator 10.1 Problem formlation This chapter concerns optimal control of dynamical systems. Most of this development concerns linear models with a particlarly simple notion of optimality.

More information

Classify by number of ports and examine the possible structures that result. Using only one-port elements, no more than two elements can be assembled.

Classify by number of ports and examine the possible structures that result. Using only one-port elements, no more than two elements can be assembled. Jnction elements in network models. Classify by nmber of ports and examine the possible strctres that reslt. Using only one-port elements, no more than two elements can be assembled. Combining two two-ports

More information

Lecture 2: August 31

Lecture 2: August 31 0-704: Information Processing and Learning Fall 206 Lecturer: Aarti Singh Lecture 2: August 3 Note: These notes are based on scribed notes from Spring5 offering of this course. LaTeX template courtesy

More information

ON THE SHAPES OF BILATERAL GAMMA DENSITIES

ON THE SHAPES OF BILATERAL GAMMA DENSITIES ON THE SHAPES OF BILATERAL GAMMA DENSITIES UWE KÜCHLER, STEFAN TAPPE Abstract. We investigate the for parameter family of bilateral Gamma distribtions. The goal of this paper is to provide a thorogh treatment

More information

RELIABILITY ASPECTS OF PROPORTIONAL MEAN RESIDUAL LIFE MODEL USING QUANTILE FUNC- TIONS

RELIABILITY ASPECTS OF PROPORTIONAL MEAN RESIDUAL LIFE MODEL USING QUANTILE FUNC- TIONS RELIABILITY ASPECTS OF PROPORTIONAL MEAN RESIDUAL LIFE MODEL USING QUANTILE FUNC- TIONS Athors: N.UNNIKRISHNAN NAIR Department of Statistics, Cochin University of Science Technology, Cochin, Kerala, INDIA

More information

On Multiobjective Duality For Variational Problems

On Multiobjective Duality For Variational Problems The Open Operational Research Jornal, 202, 6, -8 On Mltiobjective Dality For Variational Problems. Hsain *,, Bilal Ahmad 2 and Z. Jabeen 3 Open Access Department of Mathematics, Jaypee University of Engineering

More information

3 2D Elastostatic Problems in Cartesian Coordinates

3 2D Elastostatic Problems in Cartesian Coordinates D lastostatic Problems in Cartesian Coordinates Two dimensional elastostatic problems are discssed in this Chapter, that is, static problems of either plane stress or plane strain. Cartesian coordinates

More information

10-704: Information Processing and Learning Fall Lecture 24: Dec 7

10-704: Information Processing and Learning Fall Lecture 24: Dec 7 0-704: Information Processing and Learning Fall 206 Lecturer: Aarti Singh Lecture 24: Dec 7 Note: These notes are based on scribed notes from Spring5 offering of this course. LaTeX template courtesy of

More information

Non-Lecture I: Linear Programming. Th extremes of glory and of shame, Like east and west, become the same.

Non-Lecture I: Linear Programming. Th extremes of glory and of shame, Like east and west, become the same. The greatest flood has the soonest ebb; the sorest tempest the most sdden calm; the hottest love the coldest end; and from the deepest desire oftentimes enses the deadliest hate. Th extremes of glory and

More information

Chapter 3. Preferences and Utility

Chapter 3. Preferences and Utility Chapter 3 Preferences and Utilit Microeconomics stdies how individals make choices; different individals make different choices n important factor in making choices is individal s tastes or preferences

More information

Lecture 6: September 17

Lecture 6: September 17 10-725/36-725: Convex Optimization Fall 2015 Lecturer: Ryan Tibshirani Lecture 6: September 17 Scribes: Scribes: Wenjun Wang, Satwik Kottur, Zhiding Yu Note: LaTeX template courtesy of UC Berkeley EECS

More information

MAXIMUM AND ANTI-MAXIMUM PRINCIPLES FOR THE P-LAPLACIAN WITH A NONLINEAR BOUNDARY CONDITION. 1. Introduction. ν = λ u p 2 u.

MAXIMUM AND ANTI-MAXIMUM PRINCIPLES FOR THE P-LAPLACIAN WITH A NONLINEAR BOUNDARY CONDITION. 1. Introduction. ν = λ u p 2 u. 2005-Ojda International Conference on Nonlinear Analysis. Electronic Jornal of Differential Eqations, Conference 14, 2006, pp. 95 107. ISSN: 1072-6691. URL: http://ejde.math.txstate.ed or http://ejde.math.nt.ed

More information

Section 7.4: Integration of Rational Functions by Partial Fractions

Section 7.4: Integration of Rational Functions by Partial Fractions Section 7.4: Integration of Rational Fnctions by Partial Fractions This is abot as complicated as it gets. The Method of Partial Fractions Ecept for a few very special cases, crrently we have no way to

More information

Chem 4501 Introduction to Thermodynamics, 3 Credits Kinetics, and Statistical Mechanics. Fall Semester Homework Problem Set Number 10 Solutions

Chem 4501 Introduction to Thermodynamics, 3 Credits Kinetics, and Statistical Mechanics. Fall Semester Homework Problem Set Number 10 Solutions Chem 4501 Introdction to Thermodynamics, 3 Credits Kinetics, and Statistical Mechanics Fall Semester 2017 Homework Problem Set Nmber 10 Soltions 1. McQarrie and Simon, 10-4. Paraphrase: Apply Eler s theorem

More information

1 Undiscounted Problem (Deterministic)

1 Undiscounted Problem (Deterministic) Lectre 9: Linear Qadratic Control Problems 1 Undisconted Problem (Deterministic) Choose ( t ) 0 to Minimize (x trx t + tq t ) t=0 sbject to x t+1 = Ax t + B t, x 0 given. x t is an n-vector state, t a

More information

Graphs and Networks Lecture 5. PageRank. Lecturer: Daniel A. Spielman September 20, 2007

Graphs and Networks Lecture 5. PageRank. Lecturer: Daniel A. Spielman September 20, 2007 Graphs and Networks Lectre 5 PageRank Lectrer: Daniel A. Spielman September 20, 2007 5.1 Intro to PageRank PageRank, the algorithm reportedly sed by Google, assigns a nmerical rank to eery web page. More

More information

Stability of Model Predictive Control using Markov Chain Monte Carlo Optimisation

Stability of Model Predictive Control using Markov Chain Monte Carlo Optimisation Stability of Model Predictive Control sing Markov Chain Monte Carlo Optimisation Elilini Siva, Pal Golart, Jan Maciejowski and Nikolas Kantas Abstract We apply stochastic Lyapnov theory to perform stability

More information

Optimization via the Hamilton-Jacobi-Bellman Method: Theory and Applications

Optimization via the Hamilton-Jacobi-Bellman Method: Theory and Applications Optimization via the Hamilton-Jacobi-Bellman Method: Theory and Applications Navin Khaneja lectre notes taken by Christiane Koch Jne 24, 29 1 Variation yields a classical Hamiltonian system Sppose that

More information

Lecture 6: September 19

Lecture 6: September 19 36-755: Advanced Statistical Theory I Fall 2016 Lecture 6: September 19 Lecturer: Alessandro Rinaldo Scribe: YJ Choe Note: LaTeX template courtesy of UC Berkeley EECS dept. Disclaimer: These notes have

More information

i=1 y i 1fd i = dg= P N i=1 1fd i = dg.

i=1 y i 1fd i = dg= P N i=1 1fd i = dg. ECOOMETRICS II (ECO 240S) University of Toronto. Department of Economics. Winter 208 Instrctor: Victor Agirregabiria SOLUTIO TO FIAL EXAM Tesday, April 0, 208. From 9:00am-2:00pm (3 hors) ISTRUCTIOS: -

More information

B-469 Simplified Copositive and Lagrangian Relaxations for Linearly Constrained Quadratic Optimization Problems in Continuous and Binary Variables

B-469 Simplified Copositive and Lagrangian Relaxations for Linearly Constrained Quadratic Optimization Problems in Continuous and Binary Variables B-469 Simplified Copositive and Lagrangian Relaxations for Linearly Constrained Qadratic Optimization Problems in Continos and Binary Variables Naohiko Arima, Snyong Kim and Masakaz Kojima October 2012,

More information

Lecture Notes On THEORY OF COMPUTATION MODULE - 2 UNIT - 2

Lecture Notes On THEORY OF COMPUTATION MODULE - 2 UNIT - 2 BIJU PATNAIK UNIVERSITY OF TECHNOLOGY, ODISHA Lectre Notes On THEORY OF COMPUTATION MODULE - 2 UNIT - 2 Prepared by, Dr. Sbhend Kmar Rath, BPUT, Odisha. Tring Machine- Miscellany UNIT 2 TURING MACHINE

More information

1 The space of linear transformations from R n to R m :

1 The space of linear transformations from R n to R m : Math 540 Spring 20 Notes #4 Higher deriaties, Taylor s theorem The space of linear transformations from R n to R m We hae discssed linear transformations mapping R n to R m We can add sch linear transformations

More information

On averaged expected cost control as reliability for 1D ergodic diffusions

On averaged expected cost control as reliability for 1D ergodic diffusions On averaged expected cost control as reliability for 1D ergodic diffsions S.V. Anlova 5 6, H. Mai 7 8, A.Y. Veretennikov 9 10 Mon Nov 27 10:24:42 2017 Abstract For a Markov model described by a one-dimensional

More information

Second-Order Wave Equation

Second-Order Wave Equation Second-Order Wave Eqation A. Salih Department of Aerospace Engineering Indian Institte of Space Science and Technology, Thirvananthapram 3 December 016 1 Introdction The classical wave eqation is a second-order

More information

Modelling by Differential Equations from Properties of Phenomenon to its Investigation

Modelling by Differential Equations from Properties of Phenomenon to its Investigation Modelling by Differential Eqations from Properties of Phenomenon to its Investigation V. Kleiza and O. Prvinis Kanas University of Technology, Lithania Abstract The Panevezys camps of Kanas University

More information

Lecture 25: November 27

Lecture 25: November 27 10-725: Optimization Fall 2012 Lecture 25: November 27 Lecturer: Ryan Tibshirani Scribes: Matt Wytock, Supreeth Achar Note: LaTeX template courtesy of UC Berkeley EECS dept. Disclaimer: These notes have

More information

FEA Solution Procedure

FEA Solution Procedure EA Soltion Procedre (demonstrated with a -D bar element problem) MAE 5 - inite Element Analysis Several slides from this set are adapted from B.S. Altan, Michigan Technological University EA Procedre for

More information

The Dual of the Maximum Likelihood Method

The Dual of the Maximum Likelihood Method Department of Agricltral and Resorce Economics University of California, Davis The Dal of the Maximm Likelihood Method by Qirino Paris Working Paper No. 12-002 2012 Copyright @ 2012 by Qirino Paris All

More information

Multi-Voltage Floorplan Design with Optimal Voltage Assignment

Multi-Voltage Floorplan Design with Optimal Voltage Assignment Mlti-Voltage Floorplan Design with Optimal Voltage Assignment ABSTRACT Qian Zaichen Department of CSE The Chinese University of Hong Kong Shatin,N.T., Hong Kong zcqian@cse.chk.ed.hk In this paper, we stdy

More information

Information Theoretic views of Path Integral Control.

Information Theoretic views of Path Integral Control. Information Theoretic views of Path Integral Control. Evangelos A. Theodoro and Emannel Todorov Abstract We derive the connections of Path IntegralPI and Klback-LieblerKL control as presented in machine

More information

An Investigation into Estimating Type B Degrees of Freedom

An Investigation into Estimating Type B Degrees of Freedom An Investigation into Estimating Type B Degrees of H. Castrp President, Integrated Sciences Grop Jne, 00 Backgrond The degrees of freedom associated with an ncertainty estimate qantifies the amont of information

More information

Complex Variables. For ECON 397 Macroeconometrics Steve Cunningham

Complex Variables. For ECON 397 Macroeconometrics Steve Cunningham Comple Variables For ECON 397 Macroeconometrics Steve Cnningham Open Disks or Neighborhoods Deinition. The set o all points which satis the ineqalit

More information

BLOOM S TAXONOMY. Following Bloom s Taxonomy to Assess Students

BLOOM S TAXONOMY. Following Bloom s Taxonomy to Assess Students BLOOM S TAXONOMY Topic Following Bloom s Taonomy to Assess Stdents Smmary A handot for stdents to eplain Bloom s taonomy that is sed for item writing and test constrction to test stdents to see if they

More information

Lecture Notes: Finite Element Analysis, J.E. Akin, Rice University

Lecture Notes: Finite Element Analysis, J.E. Akin, Rice University 9. TRUSS ANALYSIS... 1 9.1 PLANAR TRUSS... 1 9. SPACE TRUSS... 11 9.3 SUMMARY... 1 9.4 EXERCISES... 15 9. Trss analysis 9.1 Planar trss: The differential eqation for the eqilibrim of an elastic bar (above)

More information

Characterizations of probability distributions via bivariate regression of record values

Characterizations of probability distributions via bivariate regression of record values Metrika (2008) 68:51 64 DOI 10.1007/s00184-007-0142-7 Characterizations of probability distribtions via bivariate regression of record vales George P. Yanev M. Ahsanllah M. I. Beg Received: 4 October 2006

More information

Discussion Papers Department of Economics University of Copenhagen

Discussion Papers Department of Economics University of Copenhagen Discssion Papers Department of Economics University of Copenhagen No. 10-06 Discssion of The Forward Search: Theory and Data Analysis, by Anthony C. Atkinson, Marco Riani, and Andrea Ceroli Søren Johansen,

More information

Conditions for Approaching the Origin without Intersecting the x-axis in the Liénard Plane

Conditions for Approaching the Origin without Intersecting the x-axis in the Liénard Plane Filomat 3:2 (27), 376 377 https://doi.org/.2298/fil7276a Pblished by Faclty of Sciences and Mathematics, University of Niš, Serbia Available at: http://www.pmf.ni.ac.rs/filomat Conditions for Approaching

More information

4 Exact laminar boundary layer solutions

4 Exact laminar boundary layer solutions 4 Eact laminar bondary layer soltions 4.1 Bondary layer on a flat plate (Blasis 1908 In Sec. 3, we derived the bondary layer eqations for 2D incompressible flow of constant viscosity past a weakly crved

More information

MATH2715: Statistical Methods

MATH2715: Statistical Methods MATH275: Statistical Methods Exercises VI (based on lectre, work week 7, hand in lectre Mon 4 Nov) ALL qestions cont towards the continos assessment for this modle. Q. The random variable X has a discrete

More information

Math 116 First Midterm October 14, 2009

Math 116 First Midterm October 14, 2009 Math 116 First Midterm October 14, 9 Name: EXAM SOLUTIONS Instrctor: Section: 1. Do not open this exam ntil yo are told to do so.. This exam has 1 pages inclding this cover. There are 9 problems. Note

More information

Network Coding for Multiple Unicasts: An Approach based on Linear Optimization

Network Coding for Multiple Unicasts: An Approach based on Linear Optimization Network Coding for Mltiple Unicasts: An Approach based on Linear Optimization Danail Traskov, Niranjan Ratnakar, Desmond S. Ln, Ralf Koetter, and Mriel Médard Abstract In this paper we consider the application

More information

Right Trapezoid Cover for Triangles of Perimeter Two

Right Trapezoid Cover for Triangles of Perimeter Two Kasetsart J (Nat Sci) 45 : 75-7 (0) Right Trapezoid Cover for Triangles of Perimeter Two Banyat Sroysang ABSTRACT A convex region covers a family of arcs if it contains a congrent copy of every arc in

More information

10-725/36-725: Convex Optimization Prerequisite Topics

10-725/36-725: Convex Optimization Prerequisite Topics 10-725/36-725: Convex Optimization Prerequisite Topics February 3, 2015 This is meant to be a brief, informal refresher of some topics that will form building blocks in this course. The content of the

More information

A Computational Study with Finite Element Method and Finite Difference Method for 2D Elliptic Partial Differential Equations

A Computational Study with Finite Element Method and Finite Difference Method for 2D Elliptic Partial Differential Equations Applied Mathematics, 05, 6, 04-4 Pblished Online November 05 in SciRes. http://www.scirp.org/jornal/am http://d.doi.org/0.46/am.05.685 A Comptational Stdy with Finite Element Method and Finite Difference

More information

Applying Laminar and Turbulent Flow and measuring Velocity Profile Using MATLAB

Applying Laminar and Turbulent Flow and measuring Velocity Profile Using MATLAB IOS Jornal of Mathematics (IOS-JM) e-issn: 78-578, p-issn: 319-765X. Volme 13, Isse 6 Ver. II (Nov. - Dec. 17), PP 5-59 www.iosrjornals.org Applying Laminar and Trblent Flow and measring Velocity Profile

More information

PIPELINE MECHANICAL DAMAGE CHARACTERIZATION BY MULTIPLE MAGNETIZATION LEVEL DECOUPLING

PIPELINE MECHANICAL DAMAGE CHARACTERIZATION BY MULTIPLE MAGNETIZATION LEVEL DECOUPLING PIPELINE MECHANICAL DAMAGE CHARACTERIZATION BY MULTIPLE MAGNETIZATION LEVEL DECOUPLING INTRODUCTION Richard 1. Davis & 1. Brce Nestleroth Battelle 505 King Ave Colmbs, OH 40201 Mechanical damage, cased

More information

L 1 -smoothing for the Ornstein-Uhlenbeck semigroup

L 1 -smoothing for the Ornstein-Uhlenbeck semigroup L -smoothing for the Ornstein-Uhlenbeck semigrop K. Ball, F. Barthe, W. Bednorz, K. Oleszkiewicz and P. Wolff September, 00 Abstract Given a probability density, we estimate the rate of decay of the measre

More information

Lecture 5: September 12

Lecture 5: September 12 10-725/36-725: Convex Optimization Fall 2015 Lecture 5: September 12 Lecturer: Lecturer: Ryan Tibshirani Scribes: Scribes: Barun Patra and Tyler Vuong Note: LaTeX template courtesy of UC Berkeley EECS

More information

A Characterization of the Domain of Beta-Divergence and Its Connection to Bregman Variational Model

A Characterization of the Domain of Beta-Divergence and Its Connection to Bregman Variational Model entropy Article A Characterization of the Domain of Beta-Divergence and Its Connection to Bregman Variational Model Hyenkyn Woo School of Liberal Arts, Korea University of Technology and Edcation, Cheonan

More information

STEP Support Programme. STEP III Hyperbolic Functions: Solutions

STEP Support Programme. STEP III Hyperbolic Functions: Solutions STEP Spport Programme STEP III Hyperbolic Fnctions: Soltions Start by sing the sbstittion t cosh x. This gives: sinh x cosh a cosh x cosh a sinh x t sinh x dt t dt t + ln t ln t + ln cosh a ln ln cosh

More information

Collective Inference on Markov Models for Modeling Bird Migration

Collective Inference on Markov Models for Modeling Bird Migration Collective Inference on Markov Models for Modeling Bird Migration Daniel Sheldon Cornell University dsheldon@cs.cornell.ed M. A. Saleh Elmohamed Cornell University saleh@cam.cornell.ed Dexter Kozen Cornell

More information

Simplified Identification Scheme for Structures on a Flexible Base

Simplified Identification Scheme for Structures on a Flexible Base Simplified Identification Scheme for Strctres on a Flexible Base L.M. Star California State University, Long Beach G. Mylonais University of Patras, Greece J.P. Stewart University of California, Los Angeles

More information

FEA Solution Procedure

FEA Solution Procedure EA Soltion rocedre (demonstrated with a -D bar element problem) MAE - inite Element Analysis Many slides from this set are originally from B.S. Altan, Michigan Technological U. EA rocedre for Static Analysis.

More information

Andrew W. Moore Professor School of Computer Science Carnegie Mellon University

Andrew W. Moore Professor School of Computer Science Carnegie Mellon University Spport Vector Machines Note to other teachers and sers of these slides. Andrew wold be delighted if yo fond this sorce material sefl in giving yor own lectres. Feel free to se these slides verbatim, or

More information

Cubic graphs have bounded slope parameter

Cubic graphs have bounded slope parameter Cbic graphs have bonded slope parameter B. Keszegh, J. Pach, D. Pálvölgyi, and G. Tóth Agst 25, 2009 Abstract We show that every finite connected graph G with maximm degree three and with at least one

More information

Discussion of The Forward Search: Theory and Data Analysis by Anthony C. Atkinson, Marco Riani, and Andrea Ceroli

Discussion of The Forward Search: Theory and Data Analysis by Anthony C. Atkinson, Marco Riani, and Andrea Ceroli 1 Introdction Discssion of The Forward Search: Theory and Data Analysis by Anthony C. Atkinson, Marco Riani, and Andrea Ceroli Søren Johansen Department of Economics, University of Copenhagen and CREATES,

More information

Pulses on a Struck String

Pulses on a Struck String 8.03 at ESG Spplemental Notes Plses on a Strck String These notes investigate specific eamples of transverse motion on a stretched string in cases where the string is at some time ndisplaced, bt with a

More information

Affine Invariant Total Variation Models

Affine Invariant Total Variation Models Affine Invariant Total Variation Models Helen Balinsky, Alexander Balinsky Media Technologies aboratory HP aboratories Bristol HP-7-94 Jne 6, 7* Total Variation, affine restoration, Sobolev ineqality,

More information

Assignment Fall 2014

Assignment Fall 2014 Assignment 5.086 Fall 04 De: Wednesday, 0 December at 5 PM. Upload yor soltion to corse website as a zip file YOURNAME_ASSIGNMENT_5 which incldes the script for each qestion as well as all Matlab fnctions

More information

Image and Multidimensional Signal Processing

Image and Multidimensional Signal Processing Image and Mltidimensional Signal Processing Professor William Hoff Dept of Electrical Engineering &Compter Science http://inside.mines.ed/~whoff/ Forier Transform Part : D discrete transforms 2 Overview

More information

A fundamental inverse problem in geosciences

A fundamental inverse problem in geosciences A fndamental inverse problem in geosciences Predict the vales of a spatial random field (SRF) sing a set of observed vales of the same and/or other SRFs. y i L i ( ) + v, i,..., n i ( P)? L i () : linear

More information

Introduction to Machine Learning Lecture 14. Mehryar Mohri Courant Institute and Google Research

Introduction to Machine Learning Lecture 14. Mehryar Mohri Courant Institute and Google Research Introduction to Machine Learning Lecture 14 Mehryar Mohri Courant Institute and Google Research mohri@cims.nyu.edu Density Estimation Maxent Models 2 Entropy Definition: the entropy of a random variable

More information

Convex Optimization and Support Vector Machine

Convex Optimization and Support Vector Machine Convex Optimization and Support Vector Machine Problem 0. Consider a two-class classification problem. The training data is L n = {(x 1, t 1 ),..., (x n, t n )}, where each t i { 1, 1} and x i R p. We

More information

UNCERTAINTY FOCUSED STRENGTH ANALYSIS MODEL

UNCERTAINTY FOCUSED STRENGTH ANALYSIS MODEL 8th International DAAAM Baltic Conference "INDUSTRIAL ENGINEERING - 19-1 April 01, Tallinn, Estonia UNCERTAINTY FOCUSED STRENGTH ANALYSIS MODEL Põdra, P. & Laaneots, R. Abstract: Strength analysis is a

More information

Linearly Solvable Markov Games

Linearly Solvable Markov Games Linearly Solvable Markov Games Krishnamrthy Dvijotham and mo Todorov Abstract Recent work has led to an interesting new theory of linearly solvable control, where the Bellman eqation characterizing the

More information

called the potential flow, and function φ is called the velocity potential.

called the potential flow, and function φ is called the velocity potential. J. Szantr Lectre No. 3 Potential flows 1 If the flid flow is irrotational, i.e. everwhere or almost everwhere in the field of flow there is rot 0 it means that there eists a scalar fnction ϕ,, z), sch

More information

Lecture 16: FTRL and Online Mirror Descent

Lecture 16: FTRL and Online Mirror Descent Lecture 6: FTRL and Online Mirror Descent Akshay Krishnamurthy akshay@cs.umass.edu November, 07 Recap Last time we saw two online learning algorithms. First we saw the Weighted Majority algorithm, which

More information

Research Article Permanence of a Discrete Predator-Prey Systems with Beddington-DeAngelis Functional Response and Feedback Controls

Research Article Permanence of a Discrete Predator-Prey Systems with Beddington-DeAngelis Functional Response and Feedback Controls Hindawi Pblishing Corporation Discrete Dynamics in Natre and Society Volme 2008 Article ID 149267 8 pages doi:101155/2008/149267 Research Article Permanence of a Discrete Predator-Prey Systems with Beddington-DeAngelis

More information

A Model-Free Adaptive Control of Pulsed GTAW

A Model-Free Adaptive Control of Pulsed GTAW A Model-Free Adaptive Control of Plsed GTAW F.L. Lv 1, S.B. Chen 1, and S.W. Dai 1 Institte of Welding Technology, Shanghai Jiao Tong University, Shanghai 00030, P.R. China Department of Atomatic Control,

More information

CDS 110b: Lecture 1-2 Introduction to Optimal Control

CDS 110b: Lecture 1-2 Introduction to Optimal Control CDS 110b: Lectre 1-2 Introdction to Optimal Control Richard M. Mrray 4 Janary 2006 Goals: Introdce the problem of optimal control as method of trajectory generation State the maimm principle and give eamples

More information

Execution time certification for gradient-based optimization in model predictive control

Execution time certification for gradient-based optimization in model predictive control Exection time certification for gradient-based optimization in model predictive control Giselsson, Ponts Pblished in: [Host pblication title missing] 2012 Link to pblication Citation for pblished version

More information

Lecture 5: September 15

Lecture 5: September 15 10-725/36-725: Convex Optimization Fall 2015 Lecture 5: September 15 Lecturer: Lecturer: Ryan Tibshirani Scribes: Scribes: Di Jin, Mengdi Wang, Bin Deng Note: LaTeX template courtesy of UC Berkeley EECS

More information

Geometric Image Manipulation. Lecture #4 Wednesday, January 24, 2018

Geometric Image Manipulation. Lecture #4 Wednesday, January 24, 2018 Geometric Image Maniplation Lectre 4 Wednesda, Janar 4, 08 Programming Assignment Image Maniplation: Contet To start with the obvios, an image is a D arra of piels Piel locations represent points on the

More information

Decision Oriented Bayesian Design of Experiments

Decision Oriented Bayesian Design of Experiments Decision Oriented Bayesian Design of Experiments Farminder S. Anand*, Jay H. Lee**, Matthew J. Realff*** *School of Chemical & Biomoleclar Engineering Georgia Institte of echnology, Atlanta, GA 3332 USA

More information

Discrete Applied Mathematics. The induced path function, monotonicity and betweenness

Discrete Applied Mathematics. The induced path function, monotonicity and betweenness Discrete Applied Mathematics 158 (2010) 426 433 Contents lists available at ScienceDirect Discrete Applied Mathematics jornal homepage: www.elsevier.com/locate/dam The indced path fnction, monotonicity

More information

Trace-class Monte Carlo Markov Chains for Bayesian Multivariate Linear Regression with Non-Gaussian Errors

Trace-class Monte Carlo Markov Chains for Bayesian Multivariate Linear Regression with Non-Gaussian Errors Trace-class Monte Carlo Markov Chains for Bayesian Mltivariate Linear Regression with Non-Gassian Errors Qian Qin and James P. Hobert Department of Statistics University of Florida Janary 6 Abstract Let

More information

FRTN10 Exercise 12. Synthesis by Convex Optimization

FRTN10 Exercise 12. Synthesis by Convex Optimization FRTN Exercise 2. 2. We want to design a controller C for the stable SISO process P as shown in Figre 2. sing the Yola parametrization and convex optimization. To do this, the control loop mst first be

More information

LINEAR COMBINATIONS AND SUBSPACES

LINEAR COMBINATIONS AND SUBSPACES CS131 Part II, Linear Algebra and Matrices CS131 Mathematics for Compter Scientists II Note 5 LINEAR COMBINATIONS AND SUBSPACES Linear combinations. In R 2 the vector (5, 3) can be written in the form

More information

Sufficient Optimality Condition for a Risk-Sensitive Control Problem for Backward Stochastic Differential Equations and an Application

Sufficient Optimality Condition for a Risk-Sensitive Control Problem for Backward Stochastic Differential Equations and an Application Jornal of Nmerical Mathematics and Stochastics, 9(1) : 48-6, 17 http://www.jnmas.org/jnmas9-4.pdf JNM@S Eclidean Press, LLC Online: ISSN 151-3 Sfficient Optimality Condition for a Risk-Sensitive Control

More information

FRÉCHET KERNELS AND THE ADJOINT METHOD

FRÉCHET KERNELS AND THE ADJOINT METHOD PART II FRÉCHET KERNES AND THE ADJOINT METHOD 1. Setp of the tomographic problem: Why gradients? 2. The adjoint method 3. Practical 4. Special topics (sorce imaging and time reversal) Setp of the tomographic

More information

Move Blocking Strategies in Receding Horizon Control

Move Blocking Strategies in Receding Horizon Control Move Blocking Strategies in Receding Horizon Control Raphael Cagienard, Pascal Grieder, Eric C. Kerrigan and Manfred Morari Abstract In order to deal with the comptational brden of optimal control, it

More information

On the Representation theorems of Riesz and Schwartz

On the Representation theorems of Riesz and Schwartz On the Representation theorems of Riesz and Schwartz Yves Hpperts Michel Willem Abstract We give simple proofs of the representation theorems of Riesz and Schwartz. 1 Introdction According to J.D. Gray

More information

4.2 First-Order Logic

4.2 First-Order Logic 64 First-Order Logic and Type Theory The problem can be seen in the two qestionable rles In the existential introdction, the term a has not yet been introdced into the derivation and its se can therefore

More information

Math 263 Assignment #3 Solutions. 1. A function z = f(x, y) is called harmonic if it satisfies Laplace s equation:

Math 263 Assignment #3 Solutions. 1. A function z = f(x, y) is called harmonic if it satisfies Laplace s equation: Math 263 Assignment #3 Soltions 1. A fnction z f(x, ) is called harmonic if it satisfies Laplace s eqation: 2 + 2 z 2 0 Determine whether or not the following are harmonic. (a) z x 2 + 2. We se the one-variable

More information

Model Discrimination of Polynomial Systems via Stochastic Inputs

Model Discrimination of Polynomial Systems via Stochastic Inputs Model Discrimination of Polynomial Systems via Stochastic Inpts D. Georgiev and E. Klavins Abstract Systems biologists are often faced with competing models for a given experimental system. Unfortnately,

More information

CRITERIA FOR TOEPLITZ OPERATORS ON THE SPHERE. Jingbo Xia

CRITERIA FOR TOEPLITZ OPERATORS ON THE SPHERE. Jingbo Xia CRITERIA FOR TOEPLITZ OPERATORS ON THE SPHERE Jingbo Xia Abstract. Let H 2 (S) be the Hardy space on the nit sphere S in C n. We show that a set of inner fnctions Λ is sfficient for the prpose of determining

More information

10-704: Information Processing and Learning Fall Lecture 9: Sept 28

10-704: Information Processing and Learning Fall Lecture 9: Sept 28 10-704: Information Processing and Learning Fall 2016 Lecturer: Siheng Chen Lecture 9: Sept 28 Note: These notes are based on scribed notes from Spring15 offering of this course. LaTeX template courtesy

More information