K-Lists. Anindya Sen and Corrie Scalisi. June 11, University of California, Santa Cruz. Anindya Sen and Corrie Scalisi (UCSC) K-Lists 1 / 25

Size: px
Start display at page:

Download "K-Lists. Anindya Sen and Corrie Scalisi. June 11, University of California, Santa Cruz. Anindya Sen and Corrie Scalisi (UCSC) K-Lists 1 / 25"

Transcription

1 K-Lists Anindya Sen and Corrie Scalisi University of California, Santa Cruz June 11, 2007 Anindya Sen and Corrie Scalisi (UCSC) K-Lists 1 / 25

2 Outline 1 Introduction 2 Experts 3 Noise-Free Case Deterministic Algorithm Mistake Bound 4 Randomized Algorithm Efficient Implementation 5 Deterministic Algorithm Median of Medians Proportion of Weight 6 Future Work 7 Conclusion Anindya Sen and Corrie Scalisi (UCSC) K-Lists 2 / 25

3 Problem How to combine the heads of k-lists into a prediction Online sequence of requests that index into the lists Want to minimize the number of misses (requests for indices that are not part of our prediction) Assume no two lists have an element in common x x x x Anindya Sen and Corrie Scalisi (UCSC) K-Lists 3 / 25

4 Motivation Applications in caching, ranking, etc. Combining elements in a k level virtual cache into a single real cache Designing a meta-search engine which combines the rankings of several popular search engine results Anindya Sen and Corrie Scalisi (UCSC) K-Lists 4 / 25

5 Two Lists Only need n experts for 2 list case Easy Partitions can be specified using 1 index i into list 1. The index into list 2 is just n i. Can use known algorithms such as Weighted Median, Weighted Average, etc. Anindya Sen and Corrie Scalisi (UCSC) K-Lists 5 / 25

6 K-Lists n = cache size, k = number of lists A partition is set P such that k j=1 P j = n Anindya Sen and Corrie Scalisi (UCSC) K-Lists 6 / 25

7 Outline 1 Introduction 2 Experts 3 Noise-Free Case Deterministic Algorithm Mistake Bound 4 Randomized Algorithm Efficient Implementation 5 Deterministic Algorithm Median of Medians Proportion of Weight 6 Future Work 7 Conclusion Anindya Sen and Corrie Scalisi (UCSC) K-Lists 7 / 25

8 Experts Two schemes: (n+1)*k weights one weight per partition For n = 3, k = 3 k(n + 1) weights 12 experts one weight per partition 10 experts Partitions: 003,012,021,030,102,111,120,201,210,300 Anindya Sen and Corrie Scalisi (UCSC) K-Lists 8 / 25

9 Outline 1 Introduction 2 Experts 3 Noise-Free Case Deterministic Algorithm Mistake Bound 4 Randomized Algorithm Efficient Implementation 5 Deterministic Algorithm Median of Medians Proportion of Weight 6 Future Work 7 Conclusion Anindya Sen and Corrie Scalisi (UCSC) K-Lists 9 / 25

10 Deterministic Algorithm for the Noise-Free Case m : number of cache pages with unknown status At each trial t, adversary maximizes m, algorithm minimizes m If A splits m amongst the k lists evenly, then any miss in list i reveals m k m k pages from the top of i (in cache) of pages from the bottom of j i (not in cache) m k x Anindya Sen and Corrie Scalisi (UCSC) K-Lists 10 / 25

11 Mistake Bound of Deterministic Algorithm Theorem Let M be the number of misses suffered by A for a sequence of requests for which there exists a partition with no misses. We obtain the following bound on M Proof. M 0.7k log(n) Recurrence : M(n) = M(n n + 1 ) + 1 k M(n(1 1/k)) + 1 Anindya Sen and Corrie Scalisi (UCSC) K-Lists 11 / 25

12 Mistake Bound of Deterministic Algorithm Theorem Let M be the number of misses suffered by A for a sequence of requests for which there exists a partition with no misses. We obtain the following bound on M Proof. M 0.7k log(n) Base case : M(1) = 0 Anindya Sen and Corrie Scalisi (UCSC) K-Lists 11 / 25

13 Mistake Bound of Deterministic Algorithm Theorem Let M be the number of misses suffered by A for a sequence of requests for which there exists a partition with no misses. We obtain the following bound on M Proof. M 0.7k log(n) Solving for M(n) : M(n) = k log(n)/ log( k 1 ) < k ln(2) log(n) Anindya Sen and Corrie Scalisi (UCSC) K-Lists 11 / 25

14 Outline 1 Introduction 2 Experts 3 Noise-Free Case Deterministic Algorithm Mistake Bound 4 Randomized Algorithm Efficient Implementation 5 Deterministic Algorithm Median of Medians Proportion of Weight 6 Future Work 7 Conclusion Anindya Sen and Corrie Scalisi (UCSC) K-Lists 12 / 25

15 Randomized Algorithm WMR selects a partition in each round Equal initial weights assigned to each partition WMR implicitly maintains a probability distribution p t over the set of all partitions P t 1 p t,i = e η s=1 L i Z where L i is the loss of the ith expert and Z represents normalization At the tth round, WMR randomly selects a partition based on p t + Able to get good bounds for the loss of algorithm w.r.t. loss of best in-hindsight partition E(M ALG ) M + 2M ln n + k ln n - May need to do lot of refetching Anindya Sen and Corrie Scalisi (UCSC) K-Lists 13 / 25

16 Efficient Implementation of WMR S(p, q, r) = # of partial partitions on the first p lists, with q elements and r mistakes S(p, q, r) = q S(p 1, q i, r c(p, i)) i=0 c(p, i) = total # of misses in the pth list below cutoff i Use Dynamic Programming to fill in nkt table for 1 p k, 0 q n and 1 r t. Time complexity : O(n 2 kt) Let P i = S(k, n, i)/z for 1 i t, the number of partitions making i mistakes Modify WMR to predict i proportional to P i. To obtain a corresponding partition, we can use backtracking. Anindya Sen and Corrie Scalisi (UCSC) K-Lists 14 / 25

17 Outline 1 Introduction 2 Experts 3 Noise-Free Case Deterministic Algorithm Mistake Bound 4 Randomized Algorithm Efficient Implementation 5 Deterministic Algorithm Median of Medians Proportion of Weight 6 Future Work 7 Conclusion Anindya Sen and Corrie Scalisi (UCSC) K-Lists 15 / 25

18 Deterministic Algorithm How can we balance the use of weights with the need to select only n experts for the real cache? Can we find a deterministic algorithm that gives a bound of the form a M + b, where M is the loss of the best expert? Anindya Sen and Corrie Scalisi (UCSC) K-Lists 16 / 25

19 Median of Medians Idea: Maybe the median of the total partition weight through each gap in each of the k-lists corresponds to a cache? Counterexample below β =.5 Accesses Partitions Loss Weight Anindya Sen and Corrie Scalisi (UCSC) K-Lists 17 / 25

20 Median of Medians Accesses Total Path weight through gap Medians of path weight through gaps are at (2,1,1). 4 cache pages, even though the cache should be of size 3. Anindya Sen and Corrie Scalisi (UCSC) K-Lists 18 / 25

21 Selection by Proportion of Weight Uses (n + 1) k experts model: When a miss occurs penalize experts above where loss occurred penalize same number of experts from bottoms of other lists Predict by taking W of the weight from the top of each list k 2 Anindya Sen and Corrie Scalisi (UCSC) K-Lists 19 / 25

22 Proportion of Weights Counterexample n = 3, k = 3, β =.5 and (n + 1) k weights initialized to 1 After a single miss in list 1 at index 1: W k 2 = = 2 Weights Taking 2 from the weight of each list results in the indices (2,1,1). 4 cache pages, even though the cache should be of size 3. Anindya Sen and Corrie Scalisi (UCSC) K-Lists 20 / 25

23 Outline 1 Introduction 2 Experts 3 Noise-Free Case Deterministic Algorithm Mistake Bound 4 Randomized Algorithm Efficient Implementation 5 Deterministic Algorithm Median of Medians Proportion of Weight 6 Future Work 7 Conclusion Anindya Sen and Corrie Scalisi (UCSC) K-Lists 21 / 25

24 Connection to the Metrical Task Problem Extensively studied in the Online Algorithms community Online algorithm A controls a system with n states located at points in a space with distance metric d At time t, A receives a task and a cost vector l, specifying cost of performing task in each state A tells the system to move from state i to j and pays cost d i,j + l j (d i,j = 0 if i = j) Performance comparison w.r.t. optimal offline algorithm E[cost A (σ)] a cost OPT (σ) + b Anindya Sen and Corrie Scalisi (UCSC) K-Lists 22 / 25

25 Connection to the Metrical Task Problem Theorem (Blum & Burch) Given n online algorithms for a problem that can be formulated as a Metrical Task System of diameter at most D > 0, and given ɛ < 1/4, the WMR algorithm can combine them such that on any request sequence σ it incurs expected cost at most (1 + 2ɛ)L + ( )D ln n ɛ where L is the cost of the best of the n algorithms on request sequence σ D = Size of list (dump contents of current cache and load that of another cache) What if we use the best sequence of experts (rather than a single best expert) as the comparator? Anindya Sen and Corrie Scalisi (UCSC) K-Lists 23 / 25

26 Outline 1 Introduction 2 Experts 3 Noise-Free Case Deterministic Algorithm Mistake Bound 4 Randomized Algorithm Efficient Implementation 5 Deterministic Algorithm Median of Medians Proportion of Weight 6 Future Work 7 Conclusion Anindya Sen and Corrie Scalisi (UCSC) K-Lists 24 / 25

27 Conclusion Randomized algorithm gives us a good bound but cannot make any guarantees on refetching Open Problem To find a deterministic algorithm that produces a valid cache and combines weights such that we get a good bound. Questions? Anindya Sen and Corrie Scalisi (UCSC) K-Lists 25 / 25

On-line Learning and the Metrical Task System Problem

On-line Learning and the Metrical Task System Problem Machine Learning, 39, 35 58, 2000. c 2000 Kluwer Academic Publishers. Printed in The Netherlands. On-line Learning and the Metrical Task System Problem AVRIM BLUM avrim+@cs.cmu.edu CARL BURCH cburch+@cs.cmu.edu

More information

Lecture 8: Decision-making under total uncertainty: the multiplicative weight algorithm. Lecturer: Sanjeev Arora

Lecture 8: Decision-making under total uncertainty: the multiplicative weight algorithm. Lecturer: Sanjeev Arora princeton univ. F 13 cos 521: Advanced Algorithm Design Lecture 8: Decision-making under total uncertainty: the multiplicative weight algorithm Lecturer: Sanjeev Arora Scribe: (Today s notes below are

More information

0.1 Motivating example: weighted majority algorithm

0.1 Motivating example: weighted majority algorithm princeton univ. F 16 cos 521: Advanced Algorithm Design Lecture 8: Decision-making under total uncertainty: the multiplicative weight algorithm Lecturer: Sanjeev Arora Scribe: Sanjeev Arora (Today s notes

More information

Littlestone s Dimension and Online Learnability

Littlestone s Dimension and Online Learnability Littlestone s Dimension and Online Learnability Shai Shalev-Shwartz Toyota Technological Institute at Chicago The Hebrew University Talk at UCSD workshop, February, 2009 Joint work with Shai Ben-David

More information

Learning, Games, and Networks

Learning, Games, and Networks Learning, Games, and Networks Abhishek Sinha Laboratory for Information and Decision Systems MIT ML Talk Series @CNRG December 12, 2016 1 / 44 Outline 1 Prediction With Experts Advice 2 Application to

More information

On-line Variance Minimization

On-line Variance Minimization On-line Variance Minimization Manfred Warmuth Dima Kuzmin University of California - Santa Cruz 19th Annual Conference on Learning Theory M. Warmuth, D. Kuzmin (UCSC) On-line Variance Minimization COLT06

More information

Online Kernel PCA with Entropic Matrix Updates

Online Kernel PCA with Entropic Matrix Updates Online Kernel PCA with Entropic Matrix Updates Dima Kuzmin Manfred K. Warmuth University of California - Santa Cruz ICML 2007, Corvallis, Oregon April 23, 2008 D. Kuzmin, M. Warmuth (UCSC) Online Kernel

More information

CS261: Problem Set #3

CS261: Problem Set #3 CS261: Problem Set #3 Due by 11:59 PM on Tuesday, February 23, 2016 Instructions: (1) Form a group of 1-3 students. You should turn in only one write-up for your entire group. (2) Submission instructions:

More information

1 Overview. 2 Learning from Experts. 2.1 Defining a meaningful benchmark. AM 221: Advanced Optimization Spring 2016

1 Overview. 2 Learning from Experts. 2.1 Defining a meaningful benchmark. AM 221: Advanced Optimization Spring 2016 AM 1: Advanced Optimization Spring 016 Prof. Yaron Singer Lecture 11 March 3rd 1 Overview In this lecture we will introduce the notion of online convex optimization. This is an extremely useful framework

More information

Quicksort. Where Average and Worst Case Differ. S.V. N. (vishy) Vishwanathan. University of California, Santa Cruz

Quicksort. Where Average and Worst Case Differ. S.V. N. (vishy) Vishwanathan. University of California, Santa Cruz Quicksort Where Average and Worst Case Differ S.V. N. (vishy) Vishwanathan University of California, Santa Cruz vishy@ucsc.edu February 1, 2016 S.V. N. Vishwanathan (UCSC) CMPS101 1 / 28 Basic Idea Outline

More information

A Primal-Dual Randomized Algorithm for Weighted Paging

A Primal-Dual Randomized Algorithm for Weighted Paging A Primal-Dual Randomized Algorithm for Weighted Paging Nikhil Bansal Niv Buchbinder Joseph (Seffi) Naor April 2, 2012 Abstract The study the weighted version of classic online paging problem where there

More information

Secretary Problems. Petropanagiotaki Maria. January MPLA, Algorithms & Complexity 2

Secretary Problems. Petropanagiotaki Maria. January MPLA, Algorithms & Complexity 2 January 15 2015 MPLA, Algorithms & Complexity 2 Simplest form of the problem 1 The candidates are totally ordered from best to worst with no ties. 2 The candidates arrive sequentially in random order.

More information

Design and Analysis of Algorithms

Design and Analysis of Algorithms CSE 101, Winter 2018 Design and Analysis of Algorithms Lecture 5: Divide and Conquer (Part 2) Class URL: http://vlsicad.ucsd.edu/courses/cse101-w18/ A Lower Bound on Convex Hull Lecture 4 Task: sort the

More information

An Online Algorithm for Maximizing Submodular Functions

An Online Algorithm for Maximizing Submodular Functions An Online Algorithm for Maximizing Submodular Functions Matthew Streeter December 20, 2007 CMU-CS-07-171 Daniel Golovin School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213 This research

More information

Recurrence Relations

Recurrence Relations Recurrence Relations Analysis Tools S.V. N. (vishy) Vishwanathan University of California, Santa Cruz vishy@ucsc.edu January 15, 2016 S.V. N. Vishwanathan (UCSC) CMPS101 1 / 29 Recurrences Outline 1 Recurrences

More information

Algorithms, Games, and Networks January 17, Lecture 2

Algorithms, Games, and Networks January 17, Lecture 2 Algorithms, Games, and Networks January 17, 2013 Lecturer: Avrim Blum Lecture 2 Scribe: Aleksandr Kazachkov 1 Readings for today s lecture Today s topic is online learning, regret minimization, and minimax

More information

Move from Perturbed scheme to exponential weighting average

Move from Perturbed scheme to exponential weighting average Move from Perturbed scheme to exponential weighting average Chunyang Xiao Abstract In an online decision problem, one makes decisions often with a pool of decisions sequence called experts but without

More information

Colored Bin Packing: Online Algorithms and Lower Bounds

Colored Bin Packing: Online Algorithms and Lower Bounds Noname manuscript No. (will be inserted by the editor) Colored Bin Packing: Online Algorithms and Lower Bounds Martin Böhm György Dósa Leah Epstein Jiří Sgall Pavel Veselý Received: date / Accepted: date

More information

Online Facility Location with Switching Costs

Online Facility Location with Switching Costs MSc Thesis Online Facility Location with Switching Costs Lydia Zakynthinou µπλ Graduate Program in Logic, Algorithms and Computation National and Kapodistrian University of Athens Supervised by: Dimitris

More information

Game Theory, On-line prediction and Boosting (Freund, Schapire)

Game Theory, On-line prediction and Boosting (Freund, Schapire) Game heory, On-line prediction and Boosting (Freund, Schapire) Idan Attias /4/208 INRODUCION he purpose of this paper is to bring out the close connection between game theory, on-line prediction and boosting,

More information

Lecture 2: Paging and AdWords

Lecture 2: Paging and AdWords Algoritmos e Incerteza (PUC-Rio INF2979, 2017.1) Lecture 2: Paging and AdWords March 20 2017 Lecturer: Marco Molinaro Scribe: Gabriel Homsi In this class we had a brief recap of the Ski Rental Problem

More information

Online Learning with Experts & Multiplicative Weights Algorithms

Online Learning with Experts & Multiplicative Weights Algorithms Online Learning with Experts & Multiplicative Weights Algorithms CS 159 lecture #2 Stephan Zheng April 1, 2016 Caltech Table of contents 1. Online Learning with Experts With a perfect expert Without perfect

More information

Adaptivity and Optimism: An Improved Exponentiated Gradient Algorithm

Adaptivity and Optimism: An Improved Exponentiated Gradient Algorithm Adaptivity and Optimism: An Improved Exponentiated Gradient Algorithm Jacob Steinhardt Percy Liang Stanford University {jsteinhardt,pliang}@cs.stanford.edu Jun 11, 2013 J. Steinhardt & P. Liang (Stanford)

More information

The k-server Problem and Fractional Analysis

The k-server Problem and Fractional Analysis The k-server Problem and Fractional Analysis Duru Türkoğlu November 7, 2005 Abstract The k-server problem, introduced by Manasse, McGeoch and Sleator [29, 30] is a fundamental online problem where k mobile

More information

Putting the Bayes update to sleep

Putting the Bayes update to sleep Putting the Bayes update to sleep Manfred Warmuth UCSC AMS seminar 4-13-15 Joint work with Wouter M. Koolen, Dmitry Adamskiy, Olivier Bousquet Menu How adding one line of code to the multiplicative update

More information

The Blessing and the Curse

The Blessing and the Curse The Blessing and the Curse of the Multiplicative Updates Manfred K. Warmuth University of California, Santa Cruz CMPS 272, Feb 31, 2012 Thanks to David Ilstrup and Anindya Sen for helping with the slides

More information

The Free Matrix Lunch

The Free Matrix Lunch The Free Matrix Lunch Wouter M. Koolen Wojciech Kot lowski Manfred K. Warmuth Tuesday 24 th April, 2012 Koolen, Kot lowski, Warmuth (RHUL) The Free Matrix Lunch Tuesday 24 th April, 2012 1 / 26 Introduction

More information

The Multi-Arm Bandit Framework

The Multi-Arm Bandit Framework The Multi-Arm Bandit Framework A. LAZARIC (SequeL Team @INRIA-Lille) ENS Cachan - Master 2 MVA SequeL INRIA Lille MVA-RL Course In This Lecture A. LAZARIC Reinforcement Learning Algorithms Oct 29th, 2013-2/94

More information

Introduction to Algorithms / Algorithms I Lecturer: Michael Dinitz Topic: Intro to Learning Theory Date: 12/8/16

Introduction to Algorithms / Algorithms I Lecturer: Michael Dinitz Topic: Intro to Learning Theory Date: 12/8/16 600.463 Introduction to Algorithms / Algorithms I Lecturer: Michael Dinitz Topic: Intro to Learning Theory Date: 12/8/16 25.1 Introduction Today we re going to talk about machine learning, but from an

More information

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 6: Provable Approximation via Linear Programming

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 6: Provable Approximation via Linear Programming princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 6: Provable Approximation via Linear Programming Lecturer: Matt Weinberg Scribe: Sanjeev Arora One of the running themes in this course is

More information

Approximation Algorithms for the k-set Packing Problem

Approximation Algorithms for the k-set Packing Problem Approximation Algorithms for the k-set Packing Problem Marek Cygan Institute of Informatics University of Warsaw 20th October 2016, Warszawa Marek Cygan Approximation Algorithms for the k-set Packing Problem

More information

ECE521 week 3: 23/26 January 2017

ECE521 week 3: 23/26 January 2017 ECE521 week 3: 23/26 January 2017 Outline Probabilistic interpretation of linear regression - Maximum likelihood estimation (MLE) - Maximum a posteriori (MAP) estimation Bias-variance trade-off Linear

More information

Reinforcement Learning

Reinforcement Learning Reinforcement Learning Model-Based Reinforcement Learning Model-based, PAC-MDP, sample complexity, exploration/exploitation, RMAX, E3, Bayes-optimal, Bayesian RL, model learning Vien Ngo MLR, University

More information

Online Convex Optimization Using Predictions

Online Convex Optimization Using Predictions Online Convex Optimization Using Predictions Niangjun Chen Joint work with Anish Agarwal, Lachlan Andrew, Siddharth Barman, and Adam Wierman 1 c " c " (x " ) F x " 2 c ) c ) x ) F x " x ) β x ) x " 3 F

More information

Personalized Social Recommendations Accurate or Private

Personalized Social Recommendations Accurate or Private Personalized Social Recommendations Accurate or Private Presented by: Lurye Jenny Paper by: Ashwin Machanavajjhala, Aleksandra Korolova, Atish Das Sarma Outline Introduction Motivation The model General

More information

Computer Vision Group Prof. Daniel Cremers. 6. Mixture Models and Expectation-Maximization

Computer Vision Group Prof. Daniel Cremers. 6. Mixture Models and Expectation-Maximization Prof. Daniel Cremers 6. Mixture Models and Expectation-Maximization Motivation Often the introduction of latent (unobserved) random variables into a model can help to express complex (marginal) distributions

More information

A Polylogarithmic-Competitive Algorithm for the k-server Problem

A Polylogarithmic-Competitive Algorithm for the k-server Problem A Polylogarithmic-Competitive Algorithm for the k-server Problem Nikhil Bansal Niv Buchbinder Aleksander Mądry Joseph Seffi Naor Abstract We give the first polylogarithmic-competitive randomized online

More information

Tutorial: PART 1. Online Convex Optimization, A Game- Theoretic Approach to Learning.

Tutorial: PART 1. Online Convex Optimization, A Game- Theoretic Approach to Learning. Tutorial: PART 1 Online Convex Optimization, A Game- Theoretic Approach to Learning http://www.cs.princeton.edu/~ehazan/tutorial/tutorial.htm Elad Hazan Princeton University Satyen Kale Yahoo Research

More information

Using SVD to Recommend Movies

Using SVD to Recommend Movies Michael Percy University of California, Santa Cruz Last update: December 12, 2009 Last update: December 12, 2009 1 / Outline 1 Introduction 2 Singular Value Decomposition 3 Experiments 4 Conclusion Last

More information

Online Learning and Competitive Analysis: a Unified Approach. Shahar Chen

Online Learning and Competitive Analysis: a Unified Approach. Shahar Chen Online Learning and Competitive Analysis: a Unified Approach Shahar Chen Online Learning and Competitive Analysis: a Unified Approach Research Thesis Submitted in partial fulfillment of the requirements

More information

Active Learning: Disagreement Coefficient

Active Learning: Disagreement Coefficient Advanced Course in Machine Learning Spring 2010 Active Learning: Disagreement Coefficient Handouts are jointly prepared by Shie Mannor and Shai Shalev-Shwartz In previous lectures we saw examples in which

More information

Introduction to Bandit Algorithms. Introduction to Bandit Algorithms

Introduction to Bandit Algorithms. Introduction to Bandit Algorithms Stochastic K-Arm Bandit Problem Formulation Consider K arms (actions) each correspond to an unknown distribution {ν k } K k=1 with values bounded in [0, 1]. At each time t, the agent pulls an arm I t {1,...,

More information

Recap from end of last time

Recap from end of last time Topics in Machine Learning Theory Avrim Blum 09/03/4 Lecture 3: Shifting/Sleeping Experts, the Winnow Algorithm, and L Margin Bounds Recap from end of last time RWM (multiplicative weights alg) p i = w

More information

Putting Bayes to sleep

Putting Bayes to sleep Putting Bayes to sleep Wouter M. Koolen Dmitry Adamskiy Manfred Warmuth Thursday 30 th August, 2012 Menu How a beautiful and intriguing piece of technology provides new insights in existing methods and

More information

Trade-Offs in Distributed Learning and Optimization

Trade-Offs in Distributed Learning and Optimization Trade-Offs in Distributed Learning and Optimization Ohad Shamir Weizmann Institute of Science Includes joint works with Yossi Arjevani, Nathan Srebro and Tong Zhang IHES Workshop March 2016 Distributed

More information

PetaBricks: Variable Accuracy and Online Learning

PetaBricks: Variable Accuracy and Online Learning PetaBricks: Variable Accuracy and Online Learning Jason Ansel MIT - CSAIL May 4, 2011 Jason Ansel (MIT) PetaBricks May 4, 2011 1 / 40 Outline 1 Motivating Example 2 PetaBricks Language Overview 3 Variable

More information

Online Kernel PCA with Entropic Matrix Updates

Online Kernel PCA with Entropic Matrix Updates Dima Kuzmin Manfred K. Warmuth Computer Science Department, University of California - Santa Cruz dima@cse.ucsc.edu manfred@cse.ucsc.edu Abstract A number of updates for density matrices have been developed

More information

Learning with Large Number of Experts: Component Hedge Algorithm

Learning with Large Number of Experts: Component Hedge Algorithm Learning with Large Number of Experts: Component Hedge Algorithm Giulia DeSalvo and Vitaly Kuznetsov Courant Institute March 24th, 215 1 / 3 Learning with Large Number of Experts Regret of RWM is O( T

More information

CS264: Beyond Worst-Case Analysis Lecture #4: Parameterized Analysis of Online Paging

CS264: Beyond Worst-Case Analysis Lecture #4: Parameterized Analysis of Online Paging CS264: Beyond Worst-Case Analysis Lecture #4: Parameterized Analysis of Online Paging Tim Roughgarden January 19, 2017 1 Preamble Recall our three goals for the mathematical analysis of algorithms: the

More information

The Algorithmic Foundations of Adaptive Data Analysis November, Lecture The Multiplicative Weights Algorithm

The Algorithmic Foundations of Adaptive Data Analysis November, Lecture The Multiplicative Weights Algorithm he Algorithmic Foundations of Adaptive Data Analysis November, 207 Lecture 5-6 Lecturer: Aaron Roth Scribe: Aaron Roth he Multiplicative Weights Algorithm In this lecture, we define and analyze a classic,

More information

Randomized Algorithms. Zhou Jun

Randomized Algorithms. Zhou Jun Randomized Algorithms Zhou Jun 1 Content 13.1 Contention Resolution 13.2 Global Minimum Cut 13.3 *Random Variables and Expectation 13.4 Randomized Approximation Algorithm for MAX 3- SAT 13.6 Hashing 13.7

More information

Lecture 14. Clustering, K-means, and EM

Lecture 14. Clustering, K-means, and EM Lecture 14. Clustering, K-means, and EM Prof. Alan Yuille Spring 2014 Outline 1. Clustering 2. K-means 3. EM 1 Clustering Task: Given a set of unlabeled data D = {x 1,..., x n }, we do the following: 1.

More information

Introduction to Online Algorithms

Introduction to Online Algorithms Introduction to Online Algorithms Naveen Sivadasan Indian Institute of Technology Hyderabad Online Computation In an online setting, the complete input is not known in advance. Input is a request sequence

More information

CS281B/Stat241B. Statistical Learning Theory. Lecture 14.

CS281B/Stat241B. Statistical Learning Theory. Lecture 14. CS281B/Stat241B. Statistical Learning Theory. Lecture 14. Wouter M. Koolen Convex losses Exp-concave losses Mixable losses The gradient trick Specialists 1 Menu Today we solve new online learning problems

More information

Lecture 16: Perceptron and Exponential Weights Algorithm

Lecture 16: Perceptron and Exponential Weights Algorithm EECS 598-005: Theoretical Foundations of Machine Learning Fall 2015 Lecture 16: Perceptron and Exponential Weights Algorithm Lecturer: Jacob Abernethy Scribes: Yue Wang, Editors: Weiqing Yu and Andrew

More information

Online Learning versus Offline Learning*

Online Learning versus Offline Learning* Machine Learning, 29, 45 63 (1997) c 1997 Kluwer Academic Publishers. Manufactured in The Netherlands. Online Learning versus Offline Learning* SHAI BEN-DAVID Computer Science Dept., Technion, Israel.

More information

A Second-order Bound with Excess Losses

A Second-order Bound with Excess Losses A Second-order Bound with Excess Losses Pierre Gaillard 12 Gilles Stoltz 2 Tim van Erven 3 1 EDF R&D, Clamart, France 2 GREGHEC: HEC Paris CNRS, Jouy-en-Josas, France 3 Leiden University, the Netherlands

More information

Online Learning with Feedback Graphs

Online Learning with Feedback Graphs Online Learning with Feedback Graphs Claudio Gentile INRIA and Google NY clagentile@gmailcom NYC March 6th, 2018 1 Content of this lecture Regret analysis of sequential prediction problems lying between

More information

The No-Regret Framework for Online Learning

The No-Regret Framework for Online Learning The No-Regret Framework for Online Learning A Tutorial Introduction Nahum Shimkin Technion Israel Institute of Technology Haifa, Israel Stochastic Processes in Engineering IIT Mumbai, March 2013 N. Shimkin,

More information

the Diffuse Adversary

the Diffuse Adversary Bounding the Diffuse Adversary Neal E. Young* Abstract 1.1 Results Koutsoupias and Papadimitriou recently raised the question of how well deterministic on-line paging algorithms can do against a certain

More information

EASINESS IN BANDITS. Gergely Neu. Pompeu Fabra University

EASINESS IN BANDITS. Gergely Neu. Pompeu Fabra University EASINESS IN BANDITS Gergely Neu Pompeu Fabra University EASINESS IN BANDITS Gergely Neu Pompeu Fabra University THE BANDIT PROBLEM Play for T rounds attempting to maximize rewards THE BANDIT PROBLEM Play

More information

The Online Metric Matching Problem for Doubling Metrics

The Online Metric Matching Problem for Doubling Metrics The Online Metric Matching Problem for Doubling Metrics Anupam Gupta Kevin Lewi Abstract In the online minimum-cost metric matching problem, we are given a metric space with k servers. Requests arrive

More information

Totally Corrective Boosting Algorithms that Maximize the Margin

Totally Corrective Boosting Algorithms that Maximize the Margin Totally Corrective Boosting Algorithms that Maximize the Margin Manfred K. Warmuth 1 Jun Liao 1 Gunnar Rätsch 2 1 University of California, Santa Cruz 2 Friedrich Miescher Laboratory, Tübingen, Germany

More information

Almost Tight Bounds for Reordering Buffer Management *

Almost Tight Bounds for Reordering Buffer Management * Almost Tight Bounds for Reordering Buffer Management * Anna Adamaszek Artur Czumaj Matthias Englert Harald Räcke ABSTRACT We give almost tight bounds for the online reordering buffer management problem

More information

COS 511: Theoretical Machine Learning. Lecturer: Rob Schapire Lecture 24 Scribe: Sachin Ravi May 2, 2013

COS 511: Theoretical Machine Learning. Lecturer: Rob Schapire Lecture 24 Scribe: Sachin Ravi May 2, 2013 COS 5: heoretical Machine Learning Lecturer: Rob Schapire Lecture 24 Scribe: Sachin Ravi May 2, 203 Review of Zero-Sum Games At the end of last lecture, we discussed a model for two player games (call

More information

Learning Methods for Online Prediction Problems. Peter Bartlett Statistics and EECS UC Berkeley

Learning Methods for Online Prediction Problems. Peter Bartlett Statistics and EECS UC Berkeley Learning Methods for Online Prediction Problems Peter Bartlett Statistics and EECS UC Berkeley Course Synopsis A finite comparison class: A = {1,..., m}. Converting online to batch. Online convex optimization.

More information

A New Robust Partial Least Squares Regression Method

A New Robust Partial Least Squares Regression Method A New Robust Partial Least Squares Regression Method 8th of September, 2005 Universidad Carlos III de Madrid Departamento de Estadistica Motivation Robust PLS Methods PLS Algorithm Computing Robust Variance

More information

Collaborative Topic Modeling for Recommending Scientific Articles

Collaborative Topic Modeling for Recommending Scientific Articles Collaborative Topic Modeling for Recommending Scientific Articles Chong Wang and David M. Blei Best student paper award at KDD 2011 Computer Science Department, Princeton University Presented by Tian Cao

More information

Online Interval Coloring and Variants

Online Interval Coloring and Variants Online Interval Coloring and Variants Leah Epstein 1, and Meital Levy 1 Department of Mathematics, University of Haifa, 31905 Haifa, Israel. Email: lea@math.haifa.ac.il School of Computer Science, Tel-Aviv

More information

Online Learning, Mistake Bounds, Perceptron Algorithm

Online Learning, Mistake Bounds, Perceptron Algorithm Online Learning, Mistake Bounds, Perceptron Algorithm 1 Online Learning So far the focus of the course has been on batch learning, where algorithms are presented with a sample of training data, from which

More information

CSCI-567: Machine Learning (Spring 2019)

CSCI-567: Machine Learning (Spring 2019) CSCI-567: Machine Learning (Spring 2019) Prof. Victor Adamchik U of Southern California Mar. 19, 2019 March 19, 2019 1 / 43 Administration March 19, 2019 2 / 43 Administration TA3 is due this week March

More information

A loosely Bernoulli counterexample machine

A loosely Bernoulli counterexample machine A loosely Bernoulli counterexample machine Christopher Hoffman September 7, 00 Abstract In Rudolph s paper on minimal self joinings [7] he proves that a rank one mixing transformation constructed by Ornstein

More information

Online Convex Optimization. Gautam Goel, Milan Cvitkovic, and Ellen Feldman CS 159 4/5/2016

Online Convex Optimization. Gautam Goel, Milan Cvitkovic, and Ellen Feldman CS 159 4/5/2016 Online Convex Optimization Gautam Goel, Milan Cvitkovic, and Ellen Feldman CS 159 4/5/2016 The General Setting The General Setting (Cover) Given only the above, learning isn't always possible Some Natural

More information

Exponential Family and Maximum Likelihood, Gaussian Mixture Models and the EM Algorithm. by Korbinian Schwinger

Exponential Family and Maximum Likelihood, Gaussian Mixture Models and the EM Algorithm. by Korbinian Schwinger Exponential Family and Maximum Likelihood, Gaussian Mixture Models and the EM Algorithm by Korbinian Schwinger Overview Exponential Family Maximum Likelihood The EM Algorithm Gaussian Mixture Models Exponential

More information

Proportional Share Resource Allocation Outline. Proportional Share Resource Allocation Concept

Proportional Share Resource Allocation Outline. Proportional Share Resource Allocation Concept Proportional Share Resource Allocation Outline Fluid-flow resource allocation models» Packet scheduling in a network Proportional share resource allocation models» CPU scheduling in an operating system

More information

The Weighted Majority Algorithm. Amrudin Agovic 1/24/2006 Machine Learning

The Weighted Majority Algorithm. Amrudin Agovic 1/24/2006 Machine Learning The Weighted Majority Algorithm Amrudin Agovic 1/24/2006 Machine Learning Outline The Halving Algorithm/ Motivation for WM WM WML (shifting target) WMI (infinite pool) WMG (generalized) WMC (continuous)

More information

Minimax Fixed-Design Linear Regression

Minimax Fixed-Design Linear Regression JMLR: Workshop and Conference Proceedings vol 40:1 14, 2015 Mini Fixed-Design Linear Regression Peter L. Bartlett University of California at Berkeley and Queensland University of Technology Wouter M.

More information

A note on semi-online machine covering

A note on semi-online machine covering A note on semi-online machine covering Tomáš Ebenlendr 1, John Noga 2, Jiří Sgall 1, and Gerhard Woeginger 3 1 Mathematical Institute, AS CR, Žitná 25, CZ-11567 Praha 1, The Czech Republic. Email: ebik,sgall@math.cas.cz.

More information

EM-algorithm for motif discovery

EM-algorithm for motif discovery EM-algorithm for motif discovery Xiaohui Xie University of California, Irvine EM-algorithm for motif discovery p.1/19 Position weight matrix Position weight matrix representation of a motif with width

More information

Bin packing and scheduling

Bin packing and scheduling Sanders/van Stee: Approximations- und Online-Algorithmen 1 Bin packing and scheduling Overview Bin packing: problem definition Simple 2-approximation (Next Fit) Better than 3/2 is not possible Asymptotic

More information

Math 5305 Notes. Diagnostics and Remedial Measures. Jesse Crawford. Department of Mathematics Tarleton State University

Math 5305 Notes. Diagnostics and Remedial Measures. Jesse Crawford. Department of Mathematics Tarleton State University Math 5305 Notes Diagnostics and Remedial Measures Jesse Crawford Department of Mathematics Tarleton State University (Tarleton State University) Diagnostics and Remedial Measures 1 / 44 Model Assumptions

More information

Computational Game Theory Spring Semester, 2005/6. Lecturer: Yishay Mansour Scribe: Ilan Cohen, Natan Rubin, Ophir Bleiberg*

Computational Game Theory Spring Semester, 2005/6. Lecturer: Yishay Mansour Scribe: Ilan Cohen, Natan Rubin, Ophir Bleiberg* Computational Game Theory Spring Semester, 2005/6 Lecture 5: 2-Player Zero Sum Games Lecturer: Yishay Mansour Scribe: Ilan Cohen, Natan Rubin, Ophir Bleiberg* 1 5.1 2-Player Zero Sum Games In this lecture

More information

Info-Greedy Sequential Adaptive Compressed Sensing

Info-Greedy Sequential Adaptive Compressed Sensing Info-Greedy Sequential Adaptive Compressed Sensing Yao Xie Joint work with Gabor Braun and Sebastian Pokutta Georgia Institute of Technology Presented at Allerton Conference 2014 Information sensing for

More information

A Tale of Two Metrics: Simultaneous Bounds on Competitiveness and Regret

A Tale of Two Metrics: Simultaneous Bounds on Competitiveness and Regret A Tale of Two Metrics: Simultaneous Bounds on Competitiveness and Regret Minghong Lin Computer Science California Institute of Technology Adam Wierman Computer Science California Institute of Technology

More information

Stochastic Proximal Gradient Algorithm

Stochastic Proximal Gradient Algorithm Stochastic Institut Mines-Télécom / Telecom ParisTech / Laboratoire Traitement et Communication de l Information Joint work with: Y. Atchade, Ann Arbor, USA, G. Fort LTCI/Télécom Paristech and the kind

More information

variance of independent variables: sum of variances So chebyshev predicts won t stray beyond stdev.

variance of independent variables: sum of variances So chebyshev predicts won t stray beyond stdev. Announcements No class monday. Metric embedding seminar. Review expectation notion of high probability. Markov. Today: Book 4.1, 3.3, 4.2 Chebyshev. Remind variance, standard deviation. σ 2 = E[(X µ X

More information

Distribution-specific analysis of nearest neighbor search and classification

Distribution-specific analysis of nearest neighbor search and classification Distribution-specific analysis of nearest neighbor search and classification Sanjoy Dasgupta University of California, San Diego Nearest neighbor The primeval approach to information retrieval and classification.

More information

Online Forest Density Estimation

Online Forest Density Estimation Online Forest Density Estimation Frédéric Koriche CRIL - CNRS UMR 8188, Univ. Artois koriche@cril.fr UAI 16 1 Outline 1 Probabilistic Graphical Models 2 Online Density Estimation 3 Online Forest Density

More information

An Overview of Traffic Matrix Estimation Methods

An Overview of Traffic Matrix Estimation Methods An Overview of Traffic Matrix Estimation Methods Nina Taft Berkeley www.intel.com/research Problem Statement 1 st generation solutions 2 nd generation solutions 3 rd generation solutions Summary Outline

More information

New Algorithms for Contextual Bandits

New Algorithms for Contextual Bandits New Algorithms for Contextual Bandits Lev Reyzin Georgia Institute of Technology Work done at Yahoo! 1 S A. Beygelzimer, J. Langford, L. Li, L. Reyzin, R.E. Schapire Contextual Bandit Algorithms with Supervised

More information

Distributed Machine Learning. Maria-Florina Balcan Carnegie Mellon University

Distributed Machine Learning. Maria-Florina Balcan Carnegie Mellon University Distributed Machine Learning Maria-Florina Balcan Carnegie Mellon University Distributed Machine Learning Modern applications: massive amounts of data distributed across multiple locations. Distributed

More information

Some Formal Analysis of Rocchio s Similarity-Based Relevance Feedback Algorithm

Some Formal Analysis of Rocchio s Similarity-Based Relevance Feedback Algorithm Some Formal Analysis of Rocchio s Similarity-Based Relevance Feedback Algorithm Zhixiang Chen (chen@cs.panam.edu) Department of Computer Science, University of Texas-Pan American, 1201 West University

More information

Alternatives to competitive analysis Georgios D Amanatidis

Alternatives to competitive analysis Georgios D Amanatidis Alternatives to competitive analysis Georgios D Amanatidis 1 Introduction Competitive analysis allows us to make strong theoretical statements about the performance of an algorithm without making probabilistic

More information

Divide-and-conquer: Order Statistics. Curs: Fall 2017

Divide-and-conquer: Order Statistics. Curs: Fall 2017 Divide-and-conquer: Order Statistics Curs: Fall 2017 The divide-and-conquer strategy. 1. Break the problem into smaller subproblems, 2. recursively solve each problem, 3. appropriately combine their answers.

More information

Provable Approximation via Linear Programming

Provable Approximation via Linear Programming Chapter 7 Provable Approximation via Linear Programming One of the running themes in this course is the notion of approximate solutions. Of course, this notion is tossed around a lot in applied work: whenever

More information

A Different Perspective For Approximating Max Set Packing

A Different Perspective For Approximating Max Set Packing Weizmann Institute of Science Thesis for the degree Master of Science Submitted to the Scientific Council of the Weizmann Institute of Science Rehovot, Israel A Different Perspective For Approximating

More information

Statistical Methods for Handling Incomplete Data Chapter 2: Likelihood-based approach

Statistical Methods for Handling Incomplete Data Chapter 2: Likelihood-based approach Statistical Methods for Handling Incomplete Data Chapter 2: Likelihood-based approach Jae-Kwang Kim Department of Statistics, Iowa State University Outline 1 Introduction 2 Observed likelihood 3 Mean Score

More information

Towards a Theory of Information Flow in the Finitary Process Soup

Towards a Theory of Information Flow in the Finitary Process Soup Towards a Theory of in the Finitary Process Department of Computer Science and Complexity Sciences Center University of California at Davis June 1, 2010 Goals Analyze model of evolutionary self-organization

More information

Multi-Embedding and Path Approximation of Metric Spaces

Multi-Embedding and Path Approximation of Metric Spaces Multi-Embedding and Path Approximation of Metric Spaces Yair Bartal and Manor Mendel mendelma@cshujiacil The Hebrew University Multi-Embedding and Path Approximation of Metric Spaces p1/4 Metric Embedding

More information

Rank minimization via the γ 2 norm

Rank minimization via the γ 2 norm Rank minimization via the γ 2 norm Troy Lee Columbia University Adi Shraibman Weizmann Institute Rank Minimization Problem Consider the following problem min X rank(x) A i, X b i for i = 1,..., k Arises

More information