Summer School Machine Learning Trimester



Similar documents
Machine learning challenges for big data

Workshop 1: Optimization in machine learning, vision and image processing

Big learning: challenges and opportunities

Stochastic gradient methods for machine learning

Stochastic gradient methods for machine learning

Beyond stochastic gradient descent for large-scale machine learning

Parallel & Distributed Optimization. Based on Mark Schmidt s slides

Large-Scale Similarity and Distance Metric Learning

Federated Optimization: Distributed Optimization Beyond the Datacenter

Big Data Analytics: Optimization and Randomization

Operations Research and Financial Engineering. Courses

Lecture 1: Introduction to Reinforcement Learning

TD(0) Leads to Better Policies than Approximate Value Iteration

Big Data - Lecture 1 Optimization reminders

Accelerated Parallel Optimization Methods for Large Scale Machine Learning

Machine Learning Introduction

Modern Optimization Methods for Big Data Problems MATH11146 The University of Edinburgh

350 Serra Mall, Stanford, CA

Stochastic Optimization for Big Data Analytics: Algorithms and Libraries

Winter 2016 Course Timetable. Legend: TIME: M = Monday T = Tuesday W = Wednesday R = Thursday F = Friday BREATH: M = Methodology: RA = Research Area

Professor of Mathematical Finance Office phone: Italy

Machine learning and optimization for massive data

Group Sparse Coding. Fernando Pereira Google Mountain View, CA Dennis Strelow Google Mountain View, CA

Large Scale Learning to Rank

Big Data Optimization: Randomized lock-free methods for minimizing partially separable convex functions

Parallel Data Mining. Team 2 Flash Coders Team Research Investigation Presentation 2. Foundations of Parallel Computing Oct 2014

A primal-dual algorithm for group sparse regularization with overlapping groups

Antonino Freno. Curriculum Vitae. Phone (office): Office: +33 (0)

Simple and efficient online algorithms for real world applications

Online Convex Optimization

Karthik Sridharan. 424 Gates Hall Ithaca, sridharan/ Contact Information

Motivation. Motivation. Can a software agent learn to play Backgammon by itself? Machine Learning. Reinforcement Learning

Proximal mapping via network optimization

Adaptive Online Gradient Descent

Channel Allocation in Cellular Telephone. Systems. Lab. for Info. and Decision Sciences. Cambridge, MA

CURRICULUM VITÆ. Education. Latest update : August Sanjay Ramassamy Graduate student at Brown University.

Online Learning for Matrix Factorization and Sparse Coding

Machine Learning Big Data using Map Reduce

Preliminary Syllabus for the course of Data Science for Business Analytics

Large-scale machine learning and convex optimization

Scalable Machine Learning - or what to do with all that Big Data infrastructure

Monotone multi-armed bandit allocations

A Sarsa based Autonomous Stock Trading Agent

BIOINF 585 Fall 2015 Machine Learning for Systems Biology & Clinical Informatics

Parallel Data Selection Based on Neurodynamic Optimization in the Era of Big Data

Clustering Connectionist and Statistical Language Processing

An Environment Model for N onstationary Reinforcement Learning

Master of Science in Artificial Intelligence

Big Data Optimization at SAS

Research units. Laboratory of Microbial Molecular Genetics LMGM

Neuro-Dynamic Programming An Overview

9700 South Cass Avenue, Lemont, IL URL: fulin

Foundations of Machine Learning On-Line Learning. Mehryar Mohri Courant Institute and Google Research

NEURAL NETWORKS AND REINFORCEMENT LEARNING. Abhijit Gosavi

Efficient Projections onto the l 1 -Ball for Learning in High Dimensions

SURVEY REPORT DATA SCIENCE SOCIETY 2014

Laboratoire d Informatique de Paris Nord, Institut Galilée, Université. 99 avenue Jean-Baptiste Clément, Villetaneuse, France.

How To Get A Computer Science Degree At Appalachian State

Distributed Machine Learning and Big Data

Curriculum vitae. July 2007 present Professor of Mathematics (W3), Technische

Machine Learning. CUNY Graduate Center, Spring Professor Liang Huang.

Introduction to Online Learning Theory

EIT ICT Labs MASTER SCHOOL DSS Programme Specialisations

large-scale machine learning revisited Léon Bottou Microsoft Research (NYC)

Lecture 2: The SVM classifier

Steven C.H. Hoi. School of Computer Engineering Nanyang Technological University Singapore

Ph.D., Particle Physics Theory Thesis title: FCNC Processes of B and K Mesons from Lattice QCD University of Edinburgh October 1995 July1999

Large Linear Classification When Data Cannot Fit In Memory

Beyond Reward: The Problem of Knowledge and Data

Learning the Structure of Factored Markov Decision Processes in Reinforcement Learning Problems

Dr V. J. Brown. Neuroscience (see Biomedical Sciences) History, Philosophy, Social Anthropology, Theological Studies.

Online Lazy Updates for Portfolio Selection with Transaction Costs

The Science Education System in Oxford and the UK. David Andrews

Support Vector Machines with Clustering for Training with Very Large Datasets

The Fuqua School of Business

Statistical Inference, Learning and Models for Big Data

Bayesian Machine Learning (ML): Modeling And Inference in Big Data. Zhuhua Cai Google, Rice University

Feature Selection with Monte-Carlo Tree Search

Numerical Analysis. Professor Donna Calhoun. Fall 2013 Math 465/565. Office : MG241A Office Hours : Wednesday 10:00-12:00 and 1:00-3:00

10. Proximal point method

Teaching in School of Electronic, Information and Electrical Engineering

Simon Lacoste-Julien (December 2012 version) Curriculum Vitae

Learning outcomes. Knowledge and understanding. Competence and skills

FRANCESCO BELLOCCHIO S CURRICULUM VITAE ET STUDIORUM

CSci 538 Articial Intelligence (Machine Learning and Data Analysis)

EDEN ADVANCED STRATEGIC MANAGEMENT DOCTORAL SEMINAR. When: October 19-23, 2015 Where: INSEAD Europe campus, Fontainebleau, France

Academic Calendar for Faculty

Reinforcement Learning

Linear Threshold Units

Curriculum Vitae. Rohini Kumar

Master of Artificial Intelligence

Please consult the Department of Engineering about the Computer Engineering Emphasis.

B. Delyon. Stochastic approximation with decreasing gain: convergence and asymptotic theory. Unpublished lecture notes, Université de Rennes, 2000.

An Information Retrieval using weighted Index Terms in Natural Language document collections

Contemporary Techniques for Data Mining Social Media

Online Optimization and Personalization of Teaching Sequences

Transcription:

Summer School Machine Learning Trimester Université Paul Sabatier, Toulouse September 1418, 2015

About the CIMI Machine Learning thematic trimester From September to December 2015, the International Centre for Mathematics and Computer Science (CIMI) organizes a thematic trimester on Machine Learning. The goal of this trimester is to propose a series of scientic and pedagogical events reecting common interests of the two laboratories that founded CIMI: Institut de Mathématiques de Toulouse (IMT) and Institut de Recherche en Informatique de Toulouse (IRIT). The trimester will start with a summer school and will continue with three thematic workshops. Summer school - 14th to 18th September 2015 Workshop 1 - Optimization in Machine Learning, Vision and Image Processing - 6th to 7th October 2015 Workshop 2 - Sequential Learning and Applications - 9th to 10th November 2015 Workshop 3 - Learning with Structured Data and Natural Language - 9th to 11th December 2015 The Summer School We propose four courses: Optimization in Machine Learning, Information Retrieval and Machine Learning, Reinforcement Learning and Dictionary Learning. Each course consists of three 2h lectures, and is illustrated by a hands-on computer session. In addition, an invited talk by Noah Smith will be given on Wednesday morning. For the hands-on sessions: Be aware that no computer will be provided for the hands-on sessions : participants are encouraged to bring their own laptops. The hands-on sessions on Optimization in Machine Learning, Reinforcement Learning, and Dictionary Learning will use the language julia. This free software is available for download at http://julialang.org/. Alternatively, participants will be able to run julia as a web service at https://juliabox.org/ (this requires to sign in with a Google account, no download or installation required). Code and data for the sessions are available at http://www.irit.fr/cimi-machine-learning/node/1. For the hands-on session on Information Retrieval and Machine Learning, several languages can be used, but it is recommended to install Python and IPython on your laptop: cf. http://ipython.org/install.html. Local Information All courses and hands-on sessions will take place in Amphitheater Einstein, building 3TP2, Université Paul Sabatier (campus de Rangueil). The map of the campus is available at: http://www.ups-tlse.fr/html/carte.pdf Contact aurelien.garivier@math.univ-toulouse.fr sebastien.gerchinovitz@math.univ-toulouse.fr josiane.mothe@irit.fr mathieu.serrurier@irit.fr Trimester Scientic Committee Francis Bach (Ecole Normale Supérieure), Sébastien Bubeck (Microsoft Research), Nicolo Cesa-Bianchi (Universitá degli Studi di Milano), Rémi Gribonval (lrisa Rennes), Marc Sebban (University Jean Monnet), Noah Smith (Carnegie Mellon University), Johan Suykens (KU Leuven), Marc Teboulle (Tel-Aviv University).

Schedule Monday, September 14th 9:30 Welcome Reception 10:30 A. Garivier Reinforcement Learning I 14:00 M. Melucci Information Retrieval and Machine Learning I 15:50 Coee Break 16:10 P. Richtárik Optimization in Machine Learning I Tuesday, September 15th 10:00 Coee 10:30 B. Scherrer Reinforcement Learning II 14:00 P. Richtárik Optimization in Machine Learning II 15:50 Coee Break 16:10 A. Lazaric & A. Garivier Hands-on session: Reinforcement Learning Wednesday, September 16th 8:30 N. Smith Structured Sparsity in Natural Language Processing (invited talk) 10:20 Coee Break 10:40 P. Richtárik Optimization in Machine Learning III 14:00 M. Melucci Information Retrieval and Machine Learning II 15:50 Coee Break 16:10 P. Richtárik Hands-on session: Optimization in Machine Learning Thursday, September 17th 8:30 J. Mairal Dictionary Learning I 10:20 Coee Break 10:40 B. Scherrer Reinforcement Learning III 14:00 J. Mairal Dictionary Learning II 15:50 Coee Break 16:10 J. Mairal & S. Gerchinovitz Hands-on session: Dictionary Learning Friday, September 18th 8:30 M. Melucci Information Retrieval and Machine Learning III 10:20 Coee Break 10:40 J. Mairal Dictionary Learning III 14:00 M. Melucci Hands-on session: Information Retrieval and Machine Learning 16:00 Closing Reception

Course 1: Optimization in Machine Learning, by Peter Richtárik Peter Richtárik is an Assistant Professor of Optimization at the University of Edinburgh. His research interests are in all areas of data science that intersect with optimization, including algorithms, machine learning, statistics, operations research, mathematics and high performance computing. Abstract: This course covers recent advances in scalable algorithms for convex optimization, with a particular emphasis on training (linear) predictors via the empirical risk minimization (ERM) paradigm. The material will be presented in a unied way wherever possible. Randomized, deterministic, primal, dual, accelerated, serial, parallel and distributed methods will be mentioned. The course will start in an unusual place: a concise yet powerful theory of randomized iterative methods for linear systems. While of an independent interest, this will highlight many of the algorithmic schemes and tools we shall encounter later in the course. Outline of the lectures: Lecture 1: Randomized Iterative Methods for Linear Systems (and more) [14] Lecture 2: Randomized Dual Methods [2, 3, 4, 6, 7, 10, 11, 12, 14] Lecture 3: Randomized Primal Methods [1, 5, 8, 9, 14, 15] Hands-on session: Minimizing Finite Sums via Dual-Free SDCA [15] Some references: 1. S. Shalev-Shwartz, Y. Singer, N. Srebro and A. Cotter. Pegasos: primal estimated sub-gradient solver for SVM, Mathematical Programming, 127(1), pp 330, 2011 2. P. Richtárik and M. Taká. Iteration complexity of randomized block-coordinate descent methods for minimizing a composite function, Mathematical Programming 144(2), pp 138, 2014 (arxiv:1107.2848) 3. S. Shalev-Shwartz and T. Zhang. Stochastic dual coordinate ascent methods for regularized loss minimization, Journal of Machine Learning Research 14, 567599, 2013 (arxiv:1209.1873) 4. P. Richtárik and M. Taká. Parallel coordinate descent methods for big data optimization, Mathematical Programming, 2015 (arxiv:1212.0873) 5. N. Le Roux, M. Schmidt, and F. Bach. A stochastic gradient method with an exponential convergence rate for nite training sets, NIPS 2012 6. P. Richtárik and M. Taká. Distributed coordinate descent method for learning with big data, arxiv:1310.2059, 2013 7. P. Richtárik and M. Taká. On optimal probabilities in stochastic coordinate descent methods, Optimization Letters, 2015 (arxiv:1310.3438) 8. R. Johnson and T. Zhang. Accelerating stochastic gradient descent using predictive variance reduction, NIPS, 2013. 9. J. Kone ný and P. Richtárik. Semi-stochastic gradient descent methods, arxiv:1312.1666, 2013 10. O. Fercoq and P. Richtárik Accelerated, Parallel and PROXimal coordinate descent, SIAM Journal on Optimization, 2015 (arxiv:1312.5799) 11. Z. Qu, P. Richtárik and T. Zhang. Randomized dual coordinate ascent with arbitrary sampling, NIPS, 2015 (arxiv:1411.5873) 12. Z. Qu, P. Richtárik, M. Taká and O. Fercoq. SDNA: Stochastic dual Newton ascent for empirical risk minimization, arxiv:1502.02268, 2015. 13. S. Shalev-Shwartz. SDCA without duality, arxiv:1502.06177, 2015. 14. R. Gower and P. Richtárik. Randomized iterative methods for linear systems, arxiv:1506.03296, 2015. 15. D. Csiba and P. Richtárik. Primal method for ERM with exible mini-batching schemes and non-convex losses, arxiv:1506.02227, 2015.

Course 2: Information Retrieval and Machine Learning, by Massimo Melucci Massimo Melucci has been Associate Professor in Computer Science at the Department of Information Engineering of the University of Padua, Italy, since 2001. He is on the Editorial Board of the Journal of IR and Associate Editor of Computer Science Review. His research interests are mainly in IR modeling. He is also currently investigating the intersection between IR and machine learning and the use of quantum mechanics in IR. He has been involved in EU and national research projects. Abstract: This course is an introduction to the intersection between Information Retrieval (IR) and Machine Learning (ML) models. ML has been at the basis of some IR tasks such as document ranking and relevance feedback. On the other hand IR poses new challenges to ML because of the peculiar nature of the context in which data are observed. In this course, I will introduce rst the tasks of IR and then the utilisation of some ML techniques to address these tasks. Outline of the lectures: Lecture 1: Introduction to Information Retrieval (key concepts, relevance feedback, evaluation methodology). Lecture 2: Information Retrieval Modeling (key concepts, boolean modeling, vector space modeling, relevance modeling, language modeling, evaluation results). Lecture 3: Machine Learning and Information Retrieval (key concepts, correspondence between IR and ML, features, approaches, applications, evaluation methodology). Hands-on session: Document Ranking: an Introduction. Some references: R. Baeza-Yates and B. Ribeiro-Neto. Modern Information Retrieval. Addison Wesley, New York, USA, II edition, 2010. W. Croft, D. Metzler, and T. Strohman. Search Engines: Information Retrieval in Practice. Addison Wesley, 2009. H. Li. Learning to Rank for Information Retrieval and Natural Language Processing. Morgan and Claypool, 2011. T.-Y. Liu. Learning to Rank for Information Retrieval. Springer, 2011. C. Manning, P. Raghavan, and H. Schütze. An Introduction to Information Retrieval. Cambridge University Press, 2008. G. Salton and M. McGill. Introduction to Modern Information Retrieval. McGraw-Hill, New York, NY, 1983. C. J. Van Rijsbergen. Information Retrieval. Butterworths, London, second edition, 1979.

Course 3: Reinforcement Learning, by Bruno Scherrer and Alessandro Lazaric Bruno Scherrer is a research scientist at INRIA in the project BIGS. He is a member of the Probability and Statistics Team at Institut Elie Cartan of Lorraine (IECL). His main research interests are stochastic optimal control, reinforcement learning, Markov decision processes, approximate dynamic programming, analysis of algorithms and stochastic processes. Alessandro Lazaric is a Junior Researcher (CR1) at INRIA Lille - Nord Europe in the SequeL team. Abstract: The course will cover the basic models and techniques of reinforcement learning (RL). We will begin by reviewing the Markov decision process (MDP) model used to formalize the interaction between a learning agent and an (unknown) dynamic environment. After introducing the dynamic programming techniques used to compute the exact optimal solution of an MDP known in advance, we will move to the actual learning problem where the MDP is unknown. We will introduce popular algorithms such as Q-learning and SARSA. This will lead to the analysis of two of the most important aspects of RL algorithms: how to trade o exploration and exploitation, and how to accurately approximate solutions. The core of the exploration-exploitation problem will be studied in the celebrated multi-armed bandit framework and its application to modern recommendation systems. Finally, a few examples of approximate dynamic programming will be presented together with some guarantees on their performance. The hands-on session will focus on implementing multi-armed bandit algorithms applied to the problem of policy optimization and online RL for a simple management problem. Outline of the lectures: Lecture 1: Introduction to RL: the Multi-Armed Bandit Problem. Lecture 2: Markov Decision Processes, Planning, and Dynamic Programming. Lecture 3: Approximate Solutions for Continuous MDPs. Hands-on session: Bandit Algorithms for Policy Optimization. Some references: R. Bellman, Dynamic Programming, Princeton University Press (1957). D. Bertsekas, Dynamic Programming and Optimal Control, Athena Scientic (2005). R. Sutton, Reinforcement Learning - An Introduction, MIT Press (1998). C. Szepesvari, Algorithms for Reinforcement Learning, Morgan & Claypool Publishers (2010).

Course 4: Dictionary Learning, by Julien Mairal Julien Mairal is a research scientist at INRIA in the project LEAR. He was previously a postdoctoral researcher in the statistics department of the university of California, Berkeley. Before that, he did his PhD at INRIA in the project WILLOW under the supervision of Jean Ponce and Francis Bach. He is interested in machine learning, optimization, computer vision, statistical signal and image processing, and also have some interest in bio-informatics and neurosciences. Abstract: In recent years, a large amount of multi-disciplinary research has been conducted on sparse models and their applications. In statistics and machine learning, the sparsity principle is used to perform model selectionthat is, automatically selecting a simple model among a large collection of them. In signal processing, sparse coding consists of representing data with linear combinations of a few dictionary elements. Subsequently, the corresponding tools have been widely adopted by several scientic communities such as neuroscience, bioinformatics, or computer vision. The goal of this course is to oer a self-contained view of sparse modeling for visual recognition and image processing. More specically, we focus on applications where the dictionary is learned and adapted to data, yielding a compact representation that has been successful in various contexts. Outline of the lectures: Lecture 1: A Short Introduction to Parsimony. Lecture 2: Sparse Models for Image Processing. Lecture 3: Optimization for Sparse Estimation. Hands-on session: Sparse Estimation for Image and Vision Processing. Some references: J. Mairal, F. Bach, and J. Ponce, Sparse Estimation for Image and Vision Processing, Foundations and Trends in Computer Graphics and Vision. 2014. F. Bach, R. Jenatton, J. Mairal, and G. Obozinski, Optimization with sparsity-inducing penalties, Foundations and Trends in Machine Learning, 2012.

About CIMI CIMI stands for Centre International de Mathématiques et Informatique de Toulouse and it is one of the Excellence projects selected by the ANR for the period 2012-2020. CIMI brings together the teams of the Institut de Mathématiques de Toulouse (IMT) and the Institut de Recherche en Informatique de Toulouse (IRIT). It aims at becoming an international reference in mathematics, computer science and their interactions. The program will attract high-level scientists and students from around the world. It includes actions towards attractiveness, such as Excellence Chairs for long-term visitors, grants for Doctoral and Post-Doctoral students, as well as fellowships for Master students. Attractiveness is further enhanced with thematic trimesters organized within CIMI on specic topics including courses, seminars and workshops. The innovative tools developed at CIMI will also have a strong economic impact on the region and will prot its industrial partners in major technological areas, making CIMI a strategic partner of the social and economic world.