the Information Bottleneck

Size: px
Start display at page:

Download "the Information Bottleneck"

Transcription

1 the Information Bottleneck Daniel Moyer December 10, 2017 Imaging Genetics Center/Information Science Institute University of Southern California

2 Sorry, no Neuroimaging! (at least not presented) 0

3 Instead, a deep dive into a relevant topic. 0

4 Instead, a medium dive into a relevant topic. 0

5 Rate-Distortion Theory

6 Sending a signal 1

7 Sending a signal Send: X = NEURON 1

8 Sending a signal Binary Wire Send: X = NEURON 1

9 Sending a signal N E Naïve encoding Len log symbols Send: X = NEURON Rec: X = NEURON 1

10 Sending a signal X Lossless (Huffman) encoding Len 3.75 = 22.5 symbols Send: X = NEURON Rec: X = NEURON 1

11 Sending a signal X (Example) Lossy Encoding Len 3.75 = 17.5 symbols Send: X = NEURON Rec: X = NERON 1

12 Choosing a lossy code What makes a good encoding? (According to Claude Shannon) Low Rate: We want short messages. Low Distortion: We want our messages to still make sense. Objective: Learn the best encoding p : X X. Clearly this also depends on a measure of distortion d : X X R + 2

13 Preliminaries Entropy (description length): H(X) = E p(x) [ log p(x)] = x p(x) log p(x) Mutual Information (Transmission Rate ): I(X; Y) = E p(x,y) [log Technically a bound. p(x, y) p(x)p(y) ] = H(X) H(X Y) change in desc. length = H(Y) H(Y X) 3

14 Rate Distortion Theory What makes a good encoding? (According to Claude Shannon) Low Rate: We want short messages. Low Distortion: We want our messages to still make sense. Objective: minimize p( x x) subject to I(X; X) d(x, X) < D given d : X X R + and D. 4

15 Rate Distortion Theory Rate Distortion Theorem Define the function (Shannon and Kolmogorov) R(D) = min p( x x) s.t. d(x, x) D I(X; X) as the minimum achievable rate under distortion constraint D. Then an encoding that achieves this rate is p( x x) = p( x) exp[ βd(x, x)] Z(x, β) 5

16 Rate Distortion Theory Sketch: How did we get there? R(D) = min p( x x) s.t. d(x, x) D I(X; X) original problem (1) F[p( x x)] = I(X; X) + βe p( x,x) [d(x, x)] Lagrange Mult. (2) δf = 0 Opt. Cond. (3) δp( x x) log p( x x) p( x) 0 = p(x)[log p( x x) p( x) λ(x) = βd(x, x) p(x) + βd(x, x) + λ(x) p(x) ] p( x x) = p( x) exp( βd(x, x)) exp( λ(x) p(x) ) (4) 6

17 Rate Distortion Theory X Sketch: How did we get there? R(D) = min I(X; X) original problem (1) p( x x) s.t. d(x, x) D F[p( x x)] = I(X; X) + βe p( x,x) [d(x, x)] Lagrange Mult. (2) δf = 0 Opt. Cond. (3) δp( x x) 0 = p(x)[log p( x x) λ(x) + βd(x, x) + p( x) p(x) ] log p( x x) λ(x) = βd(x, x) p( x) p(x) p( x x) = p( x) exp( βd(x, x)) exp( λ(x) p(x) ) (4) 6

18 Rate Distortion Theory Sketch: How did we get there? 1. We have one unknown p( x x) for our desiderata minimize p( x x) I(X; X) 2. We have two constraints, d(x, x) D and x p( x x) = We can relax the first constraint and form a functional F[p( x x)] = I(X; X) + βe p( x,x) [d(x, x)] 4. We know about Lagrange Multipliers. 7

19 Computational Solutions Blahut Arimoto Algorithm Iterate these two functions p t+1 ( x) = x p(x)p t ( x x) p t+1 ( x x) = p t( x) exp[ βd(x, x)] Z(x, β) Further we are guaranteed convergence as F is convex. 8

20 8

21 Information Bottleneck

22 Bottleneck Slides X = Weather in Florida Y = Price of Oranges 9

23 Bottleneck Slides X Most Relevant Parts of X X = Weather in Florida Y = Price of Oranges 9

24 More Preliminaries Entropy (description length) under X p(x): H(X) = E p(x) [ log p(x)] = x p(x) log p(x) Mutual Information (Transmission Rate bound): I(X; Y) = E p(x,y) [log p(x, y) p(x)p(y) ] Kullback Leibler Divergence (misspecified description length): D KL [p q] = E p(x) [log p(x) log q(x)] = E p [ log q(x)] H(X) }{{} Cross-entropy 10

25 Information Bottleneck What makes a good encoding? (With respect to Y) Low Rate: We want short messages. High Relevance: We want our messages to be relevant to some Y outside variable. Objective: Given L, minimize p( x x) subject to I(X; X) I( X, Y) > L 11

26 Bottleneck Slides The Information Bottleneck Define the function R(L) = min p( x x) s.t. I( x;y) L (Tishby, Pereira, and Bialek) I(X; X) as the minimum achievable rate while preserving L bits of mutual information. Then an encoding that achieves this rate has the form p( x x) = p( x) Z(x, β) exp( βd KL[ p(y x) p(y x) ]) 12

27 Rate Distortion Theory Sketch: How did we get there? 1. We have one unknown p( x x) for our desiderata minimize p( x x) I(X; X) 2. We have three constraints, I( X; Y) L, x p( x x) = 1, and p(y x) = 1. y 3. We can relax the first constraint and form a functional F[p( x x)] = I(X; X) + βi( X; Y) 4. We know about Lagrange Multipliers. 13

28 Bottleneck Computation Information Bottleneck Algorithm Iterate these three functions p t+1 ( x x) = p t+1 ( x) = x p t( x) Z(x, β) exp[ βd KL[ p(y x) p t (y x) ])] p(x)p t ( x x) p t+1 (y x) = y p(y x)p t (x x) In general, this will only converge locally, but we have a bound on the amount of information still in X but not in X about Y, given by I(X; Y). 14

29 Bottleneck parameter Not shown, optimization trajectories. 15

30 Similar in effect to Clustering (receiving partition of x), but No guarantee of disentangled representations Optimizing over p( x x) Relevance! Has similar problems to solve (e.g. choosing the size of encoding x). 16

31 Multivariate Bottleneck

32 Multivariate IB Preliminaries: I(X 1,..., X n ) = D KL [P(X 1,..., X n ) P(X 1 ) P(X n )] [ = E P log P(X ] 1,..., X n ) P(X 1 ) P(X n ) P consistent with DAG G = P = G P(X 1,..., X n ) = i P(X i Parents G (X i )) 17

33 Multivariate IB If P = G then I(X 1,..., X n ) = i I(X i ; Parents G (X i )) In general D(P P G ) = i I(X i ; NotParents G (X i ) Parents G (X i )) = I(X 1,..., X n ) I G 18

34 Multivariate IB X Y X Y X X Graph G input. Graph G output. Using L = (1 + γ)i(g input ) γi(g output ) this produces the regular IB. 19

35 Multivariate IB X Y X Y T 1 T 2 T 1 T 2 20

36 Scaling this might be hard. 20

37 Modern Bottleneck

38 Deep Variational IB Alemi et al. 2016, very similar to Achilles & Soatto

39 Deep Variational IB Main ideas of Alemi et al./achilles & Soatto 2016: 1. Deep networks are great function approximators. 22

40 Deep Variational IB Main ideas of Alemi et al./achilles & Soatto 2016: 1. Deep networks are great function approximators. 2. We want to optimize p(ˆx x), but propagating error past the stochastic loss L(x, y, p) = p(y ˆx)p(ˆx y) is hard. 22

41 Deep Variational IB Main ideas of Alemi et al./achilles & Soatto 2016: 1. Deep networks are great function approximators. 2. We want to optimize p(ˆx x), but propagating error past the stochastic loss L(x, y, p) = p(y ˆx)p(ˆx y) is hard. 3. Using technology from Variational Autoencoders, we can propagate derivatives from p(y ˆx) to p(ˆx x). (The re-parameterization trick of Kingma & Welling 2014) 22

42 Deep Variational IB Main ideas of Alemi et al./achilles & Soatto 2016: 1. Deep networks are great function approximators. 2. We want to optimize p(ˆx x), but propagating error past the stochastic loss L(x, y, p) = p(y ˆx)p(ˆx y) is hard. 3. Using technology from Variational Autoencoders, we can propagate derivatives from p(y ˆx) to p(ˆx x). (The re-parameterization trick of Kingma & Welling 2014) 4. Problems: calculating Mutual Information is actually quite difficult. Using an independent KDE estimator here, perhaps not optimal. 22

43 Other recent developments 23

44 Connections to Deep Learning Stolen directly from Tishby and Zaslavsky

45 Main Points of Tishby Claims: 1. To learn is to forget irrelevant information. 2. The layers of a deep network are (iteratively) applying a bottleneck principle. 3. The final layers should hopefully have only relevant information. 4. (Not in the paper) Backprop training produces a learning pattern w.r.t. the bottleneck objectives. 25

46 Counter Arguments (Under Review) Abstract from Anon. ICLR Submission this year. 26

47 Shannon s Warning

48 Other recent developments 27

arxiv:physics/ v1 [physics.data-an] 24 Apr 2000 Naftali Tishby, 1,2 Fernando C. Pereira, 3 and William Bialek 1

arxiv:physics/ v1 [physics.data-an] 24 Apr 2000 Naftali Tishby, 1,2 Fernando C. Pereira, 3 and William Bialek 1 The information bottleneck method arxiv:physics/0004057v1 [physics.data-an] 24 Apr 2000 Naftali Tishby, 1,2 Fernando C. Pereira, 3 and William Bialek 1 1 NEC Research Institute, 4 Independence Way Princeton,

More information

Computational Systems Biology: Biology X

Computational Systems Biology: Biology X Bud Mishra Room 1002, 715 Broadway, Courant Institute, NYU, New York, USA L#8:(November-08-2010) Cancer and Signals Outline 1 Bayesian Interpretation of Probabilities Information Theory Outline Bayesian

More information

Bioinformatics: Biology X

Bioinformatics: Biology X Bud Mishra Room 1002, 715 Broadway, Courant Institute, NYU, New York, USA Model Building/Checking, Reverse Engineering, Causality Outline 1 Bayesian Interpretation of Probabilities 2 Where (or of what)

More information

Introduction to Information Theory. Uncertainty. Entropy. Surprisal. Joint entropy. Conditional entropy. Mutual information.

Introduction to Information Theory. Uncertainty. Entropy. Surprisal. Joint entropy. Conditional entropy. Mutual information. L65 Dept. of Linguistics, Indiana University Fall 205 Information theory answers two fundamental questions in communication theory: What is the ultimate data compression? What is the transmission rate

More information

Dept. of Linguistics, Indiana University Fall 2015

Dept. of Linguistics, Indiana University Fall 2015 L645 Dept. of Linguistics, Indiana University Fall 2015 1 / 28 Information theory answers two fundamental questions in communication theory: What is the ultimate data compression? What is the transmission

More information

Optimal Deep Learning and the Information Bottleneck method

Optimal Deep Learning and the Information Bottleneck method 1 Optimal Deep Learning and the Information Bottleneck method ICRI-CI retreat, Haifa, May 2015 Naftali Tishby Noga Zaslavsky School of Engineering and Computer Science The Edmond & Lily Safra Center for

More information

Information Theory Primer:

Information Theory Primer: Information Theory Primer: Entropy, KL Divergence, Mutual Information, Jensen s inequality Seungjin Choi Department of Computer Science and Engineering Pohang University of Science and Technology 77 Cheongam-ro,

More information

Complex Systems Methods 2. Conditional mutual information, entropy rate and algorithmic complexity

Complex Systems Methods 2. Conditional mutual information, entropy rate and algorithmic complexity Complex Systems Methods 2. Conditional mutual information, entropy rate and algorithmic complexity Eckehard Olbrich MPI MiS Leipzig Potsdam WS 2007/08 Olbrich (Leipzig) 26.10.2007 1 / 18 Overview 1 Summary

More information

Information Theory. Coding and Information Theory. Information Theory Textbooks. Entropy

Information Theory. Coding and Information Theory. Information Theory Textbooks. Entropy Coding and Information Theory Chris Williams, School of Informatics, University of Edinburgh Overview What is information theory? Entropy Coding Information Theory Shannon (1948): Information theory is

More information

Lecture 22: Final Review

Lecture 22: Final Review Lecture 22: Final Review Nuts and bolts Fundamental questions and limits Tools Practical algorithms Future topics Dr Yao Xie, ECE587, Information Theory, Duke University Basics Dr Yao Xie, ECE587, Information

More information

4F5: Advanced Communications and Coding Handout 2: The Typical Set, Compression, Mutual Information

4F5: Advanced Communications and Coding Handout 2: The Typical Set, Compression, Mutual Information 4F5: Advanced Communications and Coding Handout 2: The Typical Set, Compression, Mutual Information Ramji Venkataramanan Signal Processing and Communications Lab Department of Engineering ramji.v@eng.cam.ac.uk

More information

TTIC 31230, Fundamentals of Deep Learning David McAllester, April Information Theory and Distribution Modeling

TTIC 31230, Fundamentals of Deep Learning David McAllester, April Information Theory and Distribution Modeling TTIC 31230, Fundamentals of Deep Learning David McAllester, April 2017 Information Theory and Distribution Modeling Why do we model distributions and conditional distributions using the following objective

More information

3. If a choice is broken down into two successive choices, the original H should be the weighted sum of the individual values of H.

3. If a choice is broken down into two successive choices, the original H should be the weighted sum of the individual values of H. Appendix A Information Theory A.1 Entropy Shannon (Shanon, 1948) developed the concept of entropy to measure the uncertainty of a discrete random variable. Suppose X is a discrete random variable that

More information

Deep Learning and Information Theory

Deep Learning and Information Theory Deep Learning and Information Theory Bhumesh Kumar (13D070060) Alankar Kotwal (12D070010) November 21, 2016 Abstract T he machine learning revolution has recently led to the development of a new flurry

More information

Quiz 2 Date: Monday, November 21, 2016

Quiz 2 Date: Monday, November 21, 2016 10-704 Information Processing and Learning Fall 2016 Quiz 2 Date: Monday, November 21, 2016 Name: Andrew ID: Department: Guidelines: 1. PLEASE DO NOT TURN THIS PAGE UNTIL INSTRUCTED. 2. Write your name,

More information

COMPSCI 650 Applied Information Theory Jan 21, Lecture 2

COMPSCI 650 Applied Information Theory Jan 21, Lecture 2 COMPSCI 650 Applied Information Theory Jan 21, 2016 Lecture 2 Instructor: Arya Mazumdar Scribe: Gayane Vardoyan, Jong-Chyi Su 1 Entropy Definition: Entropy is a measure of uncertainty of a random variable.

More information

Expectation Maximization

Expectation Maximization Expectation Maximization Seungjin Choi Department of Computer Science and Engineering Pohang University of Science and Technology 77 Cheongam-ro, Nam-gu, Pohang 37673, Korea seungjin@postech.ac.kr 1 /

More information

Hands-On Learning Theory Fall 2016, Lecture 3

Hands-On Learning Theory Fall 2016, Lecture 3 Hands-On Learning Theory Fall 016, Lecture 3 Jean Honorio jhonorio@purdue.edu 1 Information Theory First, we provide some information theory background. Definition 3.1 (Entropy). The entropy of a discrete

More information

Information Theory. David Rosenberg. June 15, New York University. David Rosenberg (New York University) DS-GA 1003 June 15, / 18

Information Theory. David Rosenberg. June 15, New York University. David Rosenberg (New York University) DS-GA 1003 June 15, / 18 Information Theory David Rosenberg New York University June 15, 2015 David Rosenberg (New York University) DS-GA 1003 June 15, 2015 1 / 18 A Measure of Information? Consider a discrete random variable

More information

Communication Theory and Engineering

Communication Theory and Engineering Communication Theory and Engineering Master's Degree in Electronic Engineering Sapienza University of Rome A.A. 018-019 Information theory Practice work 3 Review For any probability distribution, we define

More information

The Method of Types and Its Application to Information Hiding

The Method of Types and Its Application to Information Hiding The Method of Types and Its Application to Information Hiding Pierre Moulin University of Illinois at Urbana-Champaign www.ifp.uiuc.edu/ moulin/talks/eusipco05-slides.pdf EUSIPCO Antalya, September 7,

More information

Information Theory in Intelligent Decision Making

Information Theory in Intelligent Decision Making Information Theory in Intelligent Decision Making Adaptive Systems and Algorithms Research Groups School of Computer Science University of Hertfordshire, United Kingdom June 7, 2015 Information Theory

More information

Lecture 1: Introduction, Entropy and ML estimation

Lecture 1: Introduction, Entropy and ML estimation 0-704: Information Processing and Learning Spring 202 Lecture : Introduction, Entropy and ML estimation Lecturer: Aarti Singh Scribes: Min Xu Disclaimer: These notes have not been subjected to the usual

More information

PART III. Outline. Codes and Cryptography. Sources. Optimal Codes (I) Jorge L. Villar. MAMME, Fall 2015

PART III. Outline. Codes and Cryptography. Sources. Optimal Codes (I) Jorge L. Villar. MAMME, Fall 2015 Outline Codes and Cryptography 1 Information Sources and Optimal Codes 2 Building Optimal Codes: Huffman Codes MAMME, Fall 2015 3 Shannon Entropy and Mutual Information PART III Sources Information source:

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

Dobbiaco Lectures 2010 (4)Solved and Unsolved Problems in

Dobbiaco Lectures 2010 (4)Solved and Unsolved Problems in Dobbiaco Lectures 2010 (4) Solved and Unsolved Problems in Biology Bud Mishra Room 1002, 715 Broadway, Courant Institute, NYU, New York, USA Dobbiaco Outline 1 2 3 PART IX: Information Main theses Outline

More information

Information & Correlation

Information & Correlation Information & Correlation Jilles Vreeken 11 June 2014 (TADA) Questions of the day What is information? How can we measure correlation? and what do talking drums have to do with this? Bits and Pieces What

More information

Expectation Propagation Algorithm

Expectation Propagation Algorithm Expectation Propagation Algorithm 1 Shuang Wang School of Electrical and Computer Engineering University of Oklahoma, Tulsa, OK, 74135 Email: {shuangwang}@ou.edu This note contains three parts. First,

More information

Compressing Tabular Data via Pairwise Dependencies

Compressing Tabular Data via Pairwise Dependencies Compressing Tabular Data via Pairwise Dependencies Amir Ingber, Yahoo! Research TCE Conference, June 22, 2017 Joint work with Dmitri Pavlichin, Tsachy Weissman (Stanford) Huge datasets: everywhere - Internet

More information

How to Quantitate a Markov Chain? Stochostic project 1

How to Quantitate a Markov Chain? Stochostic project 1 How to Quantitate a Markov Chain? Stochostic project 1 Chi-Ning,Chou Wei-chang,Lee PROFESSOR RAOUL NORMAND April 18, 2015 Abstract In this project, we want to quantitatively evaluate a Markov chain. In

More information

Machine Learning Srihari. Information Theory. Sargur N. Srihari

Machine Learning Srihari. Information Theory. Sargur N. Srihari Information Theory Sargur N. Srihari 1 Topics 1. Entropy as an Information Measure 1. Discrete variable definition Relationship to Code Length 2. Continuous Variable Differential Entropy 2. Maximum Entropy

More information

Information Theory, Statistics, and Decision Trees

Information Theory, Statistics, and Decision Trees Information Theory, Statistics, and Decision Trees Léon Bottou COS 424 4/6/2010 Summary 1. Basic information theory. 2. Decision trees. 3. Information theory and statistics. Léon Bottou 2/31 COS 424 4/6/2010

More information

Information Bottleneck for Gaussian Variables

Information Bottleneck for Gaussian Variables Information Bottleneck for Gaussian Variables Gal Chechik Amir Globerson Naftali Tishby Yair Weiss {ggal,gamir,tishby,yweiss}@cs.huji.ac.il School of Computer Science and Engineering and The Interdisciplinary

More information

Lecture 1. Introduction

Lecture 1. Introduction Lecture 1. Introduction What is the course about? Logistics Questionnaire Dr. Yao Xie, ECE587, Information Theory, Duke University What is information? Dr. Yao Xie, ECE587, Information Theory, Duke University

More information

Machine Learning. Lecture 02.2: Basics of Information Theory. Nevin L. Zhang

Machine Learning. Lecture 02.2: Basics of Information Theory. Nevin L. Zhang Machine Learning Lecture 02.2: Basics of Information Theory Nevin L. Zhang lzhang@cse.ust.hk Department of Computer Science and Engineering The Hong Kong University of Science and Technology Nevin L. Zhang

More information

Chapter 2: Entropy and Mutual Information. University of Illinois at Chicago ECE 534, Natasha Devroye

Chapter 2: Entropy and Mutual Information. University of Illinois at Chicago ECE 534, Natasha Devroye Chapter 2: Entropy and Mutual Information Chapter 2 outline Definitions Entropy Joint entropy, conditional entropy Relative entropy, mutual information Chain rules Jensen s inequality Log-sum inequality

More information

QB LECTURE #4: Motif Finding

QB LECTURE #4: Motif Finding QB LECTURE #4: Motif Finding Adam Siepel Nov. 20, 2015 2 Plan for Today Probability models for binding sites Scoring and detecting binding sites De novo motif finding 3 Transcription Initiation Chromatin

More information

Chapter 2 Review of Classical Information Theory

Chapter 2 Review of Classical Information Theory Chapter 2 Review of Classical Information Theory Abstract This chapter presents a review of the classical information theory which plays a crucial role in this thesis. We introduce the various types of

More information

Variable selection and feature construction using methods related to information theory

Variable selection and feature construction using methods related to information theory Outline Variable selection and feature construction using methods related to information theory Kari 1 1 Intelligent Systems Lab, Motorola, Tempe, AZ IJCNN 2007 Outline Outline 1 Information Theory and

More information

DEEP LEARNING CHAPTER 3 PROBABILITY & INFORMATION THEORY

DEEP LEARNING CHAPTER 3 PROBABILITY & INFORMATION THEORY DEEP LEARNING CHAPTER 3 PROBABILITY & INFORMATION THEORY OUTLINE 3.1 Why Probability? 3.2 Random Variables 3.3 Probability Distributions 3.4 Marginal Probability 3.5 Conditional Probability 3.6 The Chain

More information

Information in Biology

Information in Biology Lecture 3: Information in Biology Tsvi Tlusty, tsvi@unist.ac.kr Living information is carried by molecular channels Living systems I. Self-replicating information processors Environment II. III. Evolve

More information

LECTURE 3. Last time:

LECTURE 3. Last time: LECTURE 3 Last time: Mutual Information. Convexity and concavity Jensen s inequality Information Inequality Data processing theorem Fano s Inequality Lecture outline Stochastic processes, Entropy rate

More information

Information-theoretic foundations of differential privacy

Information-theoretic foundations of differential privacy Information-theoretic foundations of differential privacy Darakhshan J. Mir Rutgers University, Piscataway NJ 08854, USA, mir@cs.rutgers.edu Abstract. We examine the information-theoretic foundations of

More information

Information Theory and Communication

Information Theory and Communication Information Theory and Communication Ritwik Banerjee rbanerjee@cs.stonybrook.edu c Ritwik Banerjee Information Theory and Communication 1/8 General Chain Rules Definition Conditional mutual information

More information

Lecture 5 - Information theory

Lecture 5 - Information theory Lecture 5 - Information theory Jan Bouda FI MU May 18, 2012 Jan Bouda (FI MU) Lecture 5 - Information theory May 18, 2012 1 / 42 Part I Uncertainty and entropy Jan Bouda (FI MU) Lecture 5 - Information

More information

Source Coding with Lists and Rényi Entropy or The Honey-Do Problem

Source Coding with Lists and Rényi Entropy or The Honey-Do Problem Source Coding with Lists and Rényi Entropy or The Honey-Do Problem Amos Lapidoth ETH Zurich October 8, 2013 Joint work with Christoph Bunte. A Task from your Spouse Using a fixed number of bits, your spouse

More information

Quantitative Biology Lecture 3

Quantitative Biology Lecture 3 23 nd Sep 2015 Quantitative Biology Lecture 3 Gurinder Singh Mickey Atwal Center for Quantitative Biology Summary Covariance, Correlation Confounding variables (Batch Effects) Information Theory Covariance

More information

Extracting relevant structures

Extracting relevant structures Chapter 5 Extracting relevant structures A key problem in understanding auditory coding is to identify the acoustic features that neurons at various levels of the system code. If we can map the relevant

More information

Information Theory and Feature Selection (Joint Informativeness and Tractability)

Information Theory and Feature Selection (Joint Informativeness and Tractability) Information Theory and Feature Selection (Joint Informativeness and Tractability) Leonidas Lefakis Zalando Research Labs 1 / 66 Dimensionality Reduction Feature Construction Construction X 1,..., X D f

More information

Chapter 9 Fundamental Limits in Information Theory

Chapter 9 Fundamental Limits in Information Theory Chapter 9 Fundamental Limits in Information Theory Information Theory is the fundamental theory behind information manipulation, including data compression and data transmission. 9.1 Introduction o For

More information

Information theory and decision tree

Information theory and decision tree Information theory and decision tree Jianxin Wu LAMDA Group National Key Lab for Novel Software Technology Nanjing University, China wujx2001@gmail.com June 14, 2018 Contents 1 Prefix code and Huffman

More information

Information Theory. M1 Informatique (parcours recherche et innovation) Aline Roumy. January INRIA Rennes 1/ 73

Information Theory. M1 Informatique (parcours recherche et innovation) Aline Roumy. January INRIA Rennes 1/ 73 1/ 73 Information Theory M1 Informatique (parcours recherche et innovation) Aline Roumy INRIA Rennes January 2018 Outline 2/ 73 1 Non mathematical introduction 2 Mathematical introduction: definitions

More information

Application of Information Theory, Lecture 7. Relative Entropy. Handout Mode. Iftach Haitner. Tel Aviv University.

Application of Information Theory, Lecture 7. Relative Entropy. Handout Mode. Iftach Haitner. Tel Aviv University. Application of Information Theory, Lecture 7 Relative Entropy Handout Mode Iftach Haitner Tel Aviv University. December 1, 2015 Iftach Haitner (TAU) Application of Information Theory, Lecture 7 December

More information

Coding for Discrete Source

Coding for Discrete Source EGR 544 Communication Theory 3. Coding for Discrete Sources Z. Aliyazicioglu Electrical and Computer Engineering Department Cal Poly Pomona Coding for Discrete Source Coding Represent source data effectively

More information

x log x, which is strictly convex, and use Jensen s Inequality:

x log x, which is strictly convex, and use Jensen s Inequality: 2. Information measures: mutual information 2.1 Divergence: main inequality Theorem 2.1 (Information Inequality). D(P Q) 0 ; D(P Q) = 0 iff P = Q Proof. Let ϕ(x) x log x, which is strictly convex, and

More information

Lecture 6: Gaussian Channels. Copyright G. Caire (Sample Lectures) 157

Lecture 6: Gaussian Channels. Copyright G. Caire (Sample Lectures) 157 Lecture 6: Gaussian Channels Copyright G. Caire (Sample Lectures) 157 Differential entropy (1) Definition 18. The (joint) differential entropy of a continuous random vector X n p X n(x) over R is: Z h(x

More information

Multimedia Communications. Mathematical Preliminaries for Lossless Compression

Multimedia Communications. Mathematical Preliminaries for Lossless Compression Multimedia Communications Mathematical Preliminaries for Lossless Compression What we will see in this chapter Definition of information and entropy Modeling a data source Definition of coding and when

More information

ELEMENT OF INFORMATION THEORY

ELEMENT OF INFORMATION THEORY History Table of Content ELEMENT OF INFORMATION THEORY O. Le Meur olemeur@irisa.fr Univ. of Rennes 1 http://www.irisa.fr/temics/staff/lemeur/ October 2010 1 History Table of Content VERSION: 2009-2010:

More information

Information in Biology

Information in Biology Information in Biology CRI - Centre de Recherches Interdisciplinaires, Paris May 2012 Information processing is an essential part of Life. Thinking about it in quantitative terms may is useful. 1 Living

More information

Ch. 8 Math Preliminaries for Lossy Coding. 8.4 Info Theory Revisited

Ch. 8 Math Preliminaries for Lossy Coding. 8.4 Info Theory Revisited Ch. 8 Math Preliminaries for Lossy Coding 8.4 Info Theory Revisited 1 Info Theory Goals for Lossy Coding Again just as for the lossless case Info Theory provides: Basis for Algorithms & Bounds on Performance

More information

PROBABILITY AND INFORMATION THEORY. Dr. Gjergji Kasneci Introduction to Information Retrieval WS

PROBABILITY AND INFORMATION THEORY. Dr. Gjergji Kasneci Introduction to Information Retrieval WS PROBABILITY AND INFORMATION THEORY Dr. Gjergji Kasneci Introduction to Information Retrieval WS 2012-13 1 Outline Intro Basics of probability and information theory Probability space Rules of probability

More information

Information Theory: Entropy, Markov Chains, and Huffman Coding

Information Theory: Entropy, Markov Chains, and Huffman Coding The University of Notre Dame A senior thesis submitted to the Department of Mathematics and the Glynn Family Honors Program Information Theory: Entropy, Markov Chains, and Huffman Coding Patrick LeBlanc

More information

Classification & Information Theory Lecture #8

Classification & Information Theory Lecture #8 Classification & Information Theory Lecture #8 Introduction to Natural Language Processing CMPSCI 585, Fall 2007 University of Massachusetts Amherst Andrew McCallum Today s Main Points Automatically categorizing

More information

INTRODUCTION TO INFORMATION THEORY

INTRODUCTION TO INFORMATION THEORY INTRODUCTION TO INFORMATION THEORY KRISTOFFER P. NIMARK These notes introduce the machinery of information theory which is a eld within applied mathematics. The material can be found in most textbooks

More information

Bayesian Machine Learning - Lecture 7

Bayesian Machine Learning - Lecture 7 Bayesian Machine Learning - Lecture 7 Guido Sanguinetti Institute for Adaptive and Neural Computation School of Informatics University of Edinburgh gsanguin@inf.ed.ac.uk March 4, 2015 Today s lecture 1

More information

Multiterminal Source Coding with an Entropy-Based Distortion Measure

Multiterminal Source Coding with an Entropy-Based Distortion Measure Multiterminal Source Coding with an Entropy-Based Distortion Measure Thomas Courtade and Rick Wesel Department of Electrical Engineering University of California, Los Angeles 4 August, 2011 IEEE International

More information

Information Theory and Statistics Lecture 2: Source coding

Information Theory and Statistics Lecture 2: Source coding Information Theory and Statistics Lecture 2: Source coding Łukasz Dębowski ldebowsk@ipipan.waw.pl Ph. D. Programme 2013/2014 Injections and codes Definition (injection) Function f is called an injection

More information

Capacity of a channel Shannon s second theorem. Information Theory 1/33

Capacity of a channel Shannon s second theorem. Information Theory 1/33 Capacity of a channel Shannon s second theorem Information Theory 1/33 Outline 1. Memoryless channels, examples ; 2. Capacity ; 3. Symmetric channels ; 4. Channel Coding ; 5. Shannon s second theorem,

More information

ECE521 Lectures 9 Fully Connected Neural Networks

ECE521 Lectures 9 Fully Connected Neural Networks ECE521 Lectures 9 Fully Connected Neural Networks Outline Multi-class classification Learning multi-layer neural networks 2 Measuring distance in probability space We learnt that the squared L2 distance

More information

Gaussian Lower Bound for the Information Bottleneck Limit

Gaussian Lower Bound for the Information Bottleneck Limit Journal of Machine Learning Research 18 (18 1-9 Submitted 7/17; Revised 11/17; Published 4/18 Gaussian Lower Bound for the Information Bottleneck Limit Amichai Painsky Naftali Tishby School of Computer

More information

Exercises with solutions (Set D)

Exercises with solutions (Set D) Exercises with solutions Set D. A fair die is rolled at the same time as a fair coin is tossed. Let A be the number on the upper surface of the die and let B describe the outcome of the coin toss, where

More information

18.2 Continuous Alphabet (discrete-time, memoryless) Channel

18.2 Continuous Alphabet (discrete-time, memoryless) Channel 0-704: Information Processing and Learning Spring 0 Lecture 8: Gaussian channel, Parallel channels and Rate-distortion theory Lecturer: Aarti Singh Scribe: Danai Koutra Disclaimer: These notes have not

More information

The binary entropy function

The binary entropy function ECE 7680 Lecture 2 Definitions and Basic Facts Objective: To learn a bunch of definitions about entropy and information measures that will be useful through the quarter, and to present some simple but

More information

Solutions to Homework Set #1 Sanov s Theorem, Rate distortion

Solutions to Homework Set #1 Sanov s Theorem, Rate distortion st Semester 00/ Solutions to Homework Set # Sanov s Theorem, Rate distortion. Sanov s theorem: Prove the simple version of Sanov s theorem for the binary random variables, i.e., let X,X,...,X n be a sequence

More information

MGMT 69000: Topics in High-dimensional Data Analysis Falll 2016

MGMT 69000: Topics in High-dimensional Data Analysis Falll 2016 MGMT 69000: Topics in High-dimensional Data Analysis Falll 2016 Lecture 14: Information Theoretic Methods Lecturer: Jiaming Xu Scribe: Hilda Ibriga, Adarsh Barik, December 02, 2016 Outline f-divergence

More information

(Classical) Information Theory II: Source coding

(Classical) Information Theory II: Source coding (Classical) Information Theory II: Source coding Sibasish Ghosh The Institute of Mathematical Sciences CIT Campus, Taramani, Chennai 600 113, India. p. 1 Abstract The information content of a random variable

More information

Solutions to Set #2 Data Compression, Huffman code and AEP

Solutions to Set #2 Data Compression, Huffman code and AEP Solutions to Set #2 Data Compression, Huffman code and AEP. Huffman coding. Consider the random variable ( ) x x X = 2 x 3 x 4 x 5 x 6 x 7 0.50 0.26 0. 0.04 0.04 0.03 0.02 (a) Find a binary Huffman code

More information

3F1: Signals and Systems INFORMATION THEORY Examples Paper Solutions

3F1: Signals and Systems INFORMATION THEORY Examples Paper Solutions Engineering Tripos Part IIA THIRD YEAR 3F: Signals and Systems INFORMATION THEORY Examples Paper Solutions. Let the joint probability mass function of two binary random variables X and Y be given in the

More information

Unraveling the mysteries of stochastic gradient descent on deep neural networks

Unraveling the mysteries of stochastic gradient descent on deep neural networks Unraveling the mysteries of stochastic gradient descent on deep neural networks Pratik Chaudhari UCLA VISION LAB 1 The question measures disagreement of predictions with ground truth Cat Dog... x = argmin

More information

Lecture 3. Mathematical methods in communication I. REMINDER. A. Convex Set. A set R is a convex set iff, x 1,x 2 R, θ, 0 θ 1, θx 1 + θx 2 R, (1)

Lecture 3. Mathematical methods in communication I. REMINDER. A. Convex Set. A set R is a convex set iff, x 1,x 2 R, θ, 0 θ 1, θx 1 + θx 2 R, (1) 3- Mathematical methods in communication Lecture 3 Lecturer: Haim Permuter Scribe: Yuval Carmel, Dima Khaykin, Ziv Goldfeld I. REMINDER A. Convex Set A set R is a convex set iff, x,x 2 R, θ, θ, θx + θx

More information

Representation. Stefano Ermon, Aditya Grover. Stanford University. Lecture 2

Representation. Stefano Ermon, Aditya Grover. Stanford University. Lecture 2 Representation Stefano Ermon, Aditya Grover Stanford University Lecture 2 Stefano Ermon, Aditya Grover (AI Lab) Deep Generative Models Lecture 2 1 / 32 Learning a generative model We are given a training

More information

Quantitative Biology II Lecture 4: Variational Methods

Quantitative Biology II Lecture 4: Variational Methods 10 th March 2015 Quantitative Biology II Lecture 4: Variational Methods Gurinder Singh Mickey Atwal Center for Quantitative Biology Cold Spring Harbor Laboratory Image credit: Mike West Summary Approximate

More information

Remote Source Coding with Two-Sided Information

Remote Source Coding with Two-Sided Information Remote Source Coding with Two-Sided Information Basak Guler Ebrahim MolavianJazi Aylin Yener Wireless Communications and Networking Laboratory Department of Electrical Engineering The Pennsylvania State

More information

APC486/ELE486: Transmission and Compression of Information. Bounds on the Expected Length of Code Words

APC486/ELE486: Transmission and Compression of Information. Bounds on the Expected Length of Code Words APC486/ELE486: Transmission and Compression of Information Bounds on the Expected Length of Code Words Scribe: Kiran Vodrahalli September 8, 204 Notations In these notes, denotes a finite set, called the

More information

Information Geometric view of Belief Propagation

Information Geometric view of Belief Propagation Information Geometric view of Belief Propagation Yunshu Liu 2013-10-17 References: [1]. Shiro Ikeda, Toshiyuki Tanaka and Shun-ichi Amari, Stochastic reasoning, Free energy and Information Geometry, Neural

More information

Homework 1 Due: Thursday 2/5/2015. Instructions: Turn in your homework in class on Thursday 2/5/2015

Homework 1 Due: Thursday 2/5/2015. Instructions: Turn in your homework in class on Thursday 2/5/2015 10-704 Homework 1 Due: Thursday 2/5/2015 Instructions: Turn in your homework in class on Thursday 2/5/2015 1. Information Theory Basics and Inequalities C&T 2.47, 2.29 (a) A deck of n cards in order 1,

More information

13: Variational inference II

13: Variational inference II 10-708: Probabilistic Graphical Models, Spring 2015 13: Variational inference II Lecturer: Eric P. Xing Scribes: Ronghuo Zheng, Zhiting Hu, Yuntian Deng 1 Introduction We started to talk about variational

More information

Bayesian Inference Course, WTCN, UCL, March 2013

Bayesian Inference Course, WTCN, UCL, March 2013 Bayesian Course, WTCN, UCL, March 2013 Shannon (1948) asked how much information is received when we observe a specific value of the variable x? If an unlikely event occurs then one would expect the information

More information

Causal Modeling with Generative Neural Networks

Causal Modeling with Generative Neural Networks Causal Modeling with Generative Neural Networks Michele Sebag TAO, CNRS INRIA LRI Université Paris-Sud Joint work: D. Kalainathan, O. Goudet, I. Guyon, M. Hajaiej, A. Decelle, C. Furtlehner https://arxiv.org/abs/1709.05321

More information

Chapter 4. Data Transmission and Channel Capacity. Po-Ning Chen, Professor. Department of Communications Engineering. National Chiao Tung University

Chapter 4. Data Transmission and Channel Capacity. Po-Ning Chen, Professor. Department of Communications Engineering. National Chiao Tung University Chapter 4 Data Transmission and Channel Capacity Po-Ning Chen, Professor Department of Communications Engineering National Chiao Tung University Hsin Chu, Taiwan 30050, R.O.C. Principle of Data Transmission

More information

EE376A: Homework #3 Due by 11:59pm Saturday, February 10th, 2018

EE376A: Homework #3 Due by 11:59pm Saturday, February 10th, 2018 Please submit the solutions on Gradescope. EE376A: Homework #3 Due by 11:59pm Saturday, February 10th, 2018 1. Optimal codeword lengths. Although the codeword lengths of an optimal variable length code

More information

Series 7, May 22, 2018 (EM Convergence)

Series 7, May 22, 2018 (EM Convergence) Exercises Introduction to Machine Learning SS 2018 Series 7, May 22, 2018 (EM Convergence) Institute for Machine Learning Dept. of Computer Science, ETH Zürich Prof. Dr. Andreas Krause Web: https://las.inf.ethz.ch/teaching/introml-s18

More information

Block 2: Introduction to Information Theory

Block 2: Introduction to Information Theory Block 2: Introduction to Information Theory Francisco J. Escribano April 26, 2015 Francisco J. Escribano Block 2: Introduction to Information Theory April 26, 2015 1 / 51 Table of contents 1 Motivation

More information

Optimal predictive inference

Optimal predictive inference Optimal predictive inference Susanne Still University of Hawaii at Manoa Information and Computer Sciences S. Still, J. P. Crutchfield. Structure or Noise? http://lanl.arxiv.org/abs/0708.0654 S. Still,

More information

3F1 Information Theory, Lecture 1

3F1 Information Theory, Lecture 1 3F1 Information Theory, Lecture 1 Jossy Sayir Department of Engineering Michaelmas 2013, 22 November 2013 Organisation History Entropy Mutual Information 2 / 18 Course Organisation 4 lectures Course material:

More information

Medical Imaging. Norbert Schuff, Ph.D. Center for Imaging of Neurodegenerative Diseases

Medical Imaging. Norbert Schuff, Ph.D. Center for Imaging of Neurodegenerative Diseases Uses of Information Theory in Medical Imaging Norbert Schuff, Ph.D. Center for Imaging of Neurodegenerative Diseases Norbert.schuff@ucsf.edu With contributions from Dr. Wang Zhang Medical Imaging Informatics,

More information

Machine Learning Basics: Maximum Likelihood Estimation

Machine Learning Basics: Maximum Likelihood Estimation Machine Learning Basics: Maximum Likelihood Estimation Sargur N. srihari@cedar.buffalo.edu This is part of lecture slides on Deep Learning: http://www.cedar.buffalo.edu/~srihari/cse676 1 Topics 1. Learning

More information

DATA MINING LECTURE 9. Minimum Description Length Information Theory Co-Clustering

DATA MINING LECTURE 9. Minimum Description Length Information Theory Co-Clustering DATA MINING LECTURE 9 Minimum Description Length Information Theory Co-Clustering MINIMUM DESCRIPTION LENGTH Occam s razor Most data mining tasks can be described as creating a model for the data E.g.,

More information

ECE Information theory Final

ECE Information theory Final ECE 776 - Information theory Final Q1 (1 point) We would like to compress a Gaussian source with zero mean and variance 1 We consider two strategies In the first, we quantize with a step size so that the

More information

Chaos, Complexity, and Inference (36-462)

Chaos, Complexity, and Inference (36-462) Chaos, Complexity, and Inference (36-462) Lecture 7: Information Theory Cosma Shalizi 3 February 2009 Entropy and Information Measuring randomness and dependence in bits The connection to statistics Long-run

More information