Smooth Games Optimization and Machine Learning Workshop

Constantinos Daskalakis (MIT)

Title:

Improving Generative Adversarial Networks using Game Theory and Statistics

Abstract:

Generative Adversarial Networks (aka GANs) are a recently proposed approach for learning samplers of high-dimensional distributions with intricate structure, such as distributions over natural images, given samples from these distributions. They are trained by setting up a two-player zero-sum game between two neural networks, which learn statistics of a target distribution by adapting their strategies in the game using gradient descent. Despite their intriguing performance in practice, GANs pose great challenges to both Optimization and Statistics. Their training suffers from oscillations, and they are difficult to scale to high-dimensional settings. We study how game-theoretic and statistical techniques can be brought to bare on these important challenges. We use Game Theory towards improving GAN training, and Statistics towards scaling up the dimensionality of the generated distributions.

Short Bio:

Constantinos Daskalakis works on computation theory and its interface with game theory, economics, probability theory, statistics and machine learning. His work has resolved long-standing open problems about the computational complexity of Nash equilibrium, the mathematical structure and computational complexity of multi-item auctions, and the behavior of machine-learning methods such as the expectation-maximization algorithm. He has obtained computationally and statistically efficient methods for statistical hypothesis testing and learning in high-dimensional settings, as well as results characterizing the structure and concentration properties of high-dimensional distributions. For his work on the complexity of Nash equilibrium, he has been awarded the Kalai prize from the Game Theory Society and the SIAM outstanding paper prize. He is also a recipient of the ACM doctoral dissertation award, a Simons investigator award, and the Rolf Nevanlinna Prize, one of the most prestigious international awards in mathematics.

Niao He (UIUC)

Title:

Smooth Games in Machine Learning Beyond GANs

Abstract:

This talk will discuss a wide spectrum of recent advances in machine learning using smooth games, beyond the phenomenal GANs. Such showcases include reinforcement learning, robust and adversarial machine learning, approximate Bayesian computation, maximum likelihood estimation in exponential family, and etc. We show that all of these classical machine learning tasks can be reduced to solving (non) convex-concave min-max optimization problems. Hence, it is of paramount importance to developing a good theoretical understanding and principled algorithms for min-max optimization. We will review some of the theory and algorithms for smooth games and variational inequalities in the convex regime and shed some light on their counterparts in the non-convex regime.

Short Bio:

Niao He is an assistant professor in the Department of Industrial and Enterprise Systems Engineering and Coordinated Science Laboratory at the University of Illinois at Urbana-Champaign. Before joining Illinois, she received her Ph.D. degree in Operations Research from Georgia Institute of Technology in 2015 and B.S. degree in Mathematics from University of Science and Technology of China in 2010. Her research interests are in large-scale optimization and machine learning, with a primary focus in bridging modern optimization theory and algorithms with core machine learning topics, like Bayesian inference, reinforcement learning, and adversarial learning. She is also a recipient of the NSF CISE Research Initiation Initiative (CRII) Award and the NCSA Faculty Fellowship.

Jacob Abernethy (Georgia Tech.)

Title:

Building Algorithms by Playing Games

Abstract:

A very popular trick for solving certain types of optimization problems is this: write your objective as the solution of a two-player zero-sum game, endow both players with an appropriate learning algorithm, watch how the opponents compete, and extract an (approximate) solution from the actions/decisions taken by the players throughout the process. This approach is very generic and provides a natural template to produce new and interesting algorithms. I will describe this framework and show how it applies in several scenarios, and describe recent work that draws a connection to the Frank-Wolfe algorithm and Nesterov's Accelerated Gradient Descent.

Short Bio:

Jacob Abernethy is an Assistant Professor in Computer Science at Georgia Tech. He started his faculty career in the Department of Electrical Engineering and Computer Science at the University of Michigan. In October 2011 he finished a PhD in the Division of Computer Science at the University of California at Berkeley, and then spent nearly two years as a Simons postdoctoral fellow at the CIS department at UPenn, working with Michael Kearns. Abernethy's primary interest is in Machine Learning, with a particular focus in sequential decision making, online learning, online algorithms and adversarial learning models. He did his Master's degree at TTI-C, and his Bachelor's Degree at MIT. Abernethy's PhD advisor is Prof. Peter Bartlett.

Paulina Grnarova (ETH)

Title:

An interpretation of GANs via online learning and game theory

Abstract:

Generative Adversarial Networks (GANs) have become one of the most powerful paradigms in learning real-world distributions. Despite this success, their minimax nature makes them fundamentally different to more classical generative models thus raising novel challenges; most notably in terms of training and evaluation. Indeed, finding a saddle-point is in general a harder task than converging to an extremum. We view the problem of training GANs as finding a mixed strategy in a zero-sum game. Building upon ideas from online learning and game theory, we propose (i) a novel training method with provable convergence to an equilibrium for semi-shallow GAN architectures, i.e. architectures where the discriminator is a one layer network and the generator is an arbitrary network and (ii) a natural metric for detecting non-convergence, namely the duality gap.

Short Bio:

Paulina Grnarova is a PhD student at ETH Zurich, supervised by Prof. Thomas Hofmann. Prior to her PhD she did her Master studies at EPF Lausanne. Her research mainly focuses on the development and understanding of generative models for complex data. Recently she proposed an online learning inspired algorithm for training GANs with better convergence guarantees. Since 2016, she is also a Google AI collaborator and works on improving text generation as part of a research team.

Simon Lacoste-Julien (Mila & University of Montreal)

Simon Lacoste-Julien is a CIFAR fellow and an assistant professor at Mila and DIRO from Université de Montréal. His research interests are machine learning and applied math, with applications to computer vision and natural language processing. He obtained a B.Sc. in math., physics and computer science from McGill, a PhD in computer science from UC Berkeley and a post-doc from the University of Cambridge. He spent a few years as a research faculty at INRIA and École normale supérieure in Paris before coming back to his roots in Montreal in 2016.
Simon published several papers at the intersection of mathematical programming and machine learning, and in particular for solving min-max games. He is a frequent participant to the NeurIPS OPT workshop series, and co-organized the NeurIPS 2009 workshop on “The Generative & Discriminative Learning Interface” .

Ioannis Mitliagkas (Mila & University of Montreal)

Ioannis Mitliagkas is an assistant professor in the department of Computer Science and Operations Research (DIRO) at the University of Montréal. Before that, he was a Postdoctoral Scholar with the departments of Statistics and Computer Science at Stanford University. He obtained his Ph.D. from the department of Electrical and Computer Engineering at The University of Texas at Austin. His research includes topics in optimization, statistical learning and inference, and efficient large-scale and distributed algorithms.
He is particularly interested in the dynamics of optimization, like momentum methods, in the presence of system dynamics, adaptivity, and lately, smooth two-player games (ongoing work).

Gauthier Gidel (Mila & University of Montreal)

Gauthier Gidel received the Diplôme de l’École Normale Supérieure in 2017 (ULM MPI2013) and the Master of Science MVA from École Normale supérieur Paris-Saclay in 2016. Gauthier is currently pursuing his PhD at Mila and DIRO from Université de Montréal under the supervision of Simon Lacoste-Julien.
Gauthier’s PhD thesis topic revolves around saddle point optimization (a.k.a mini-max problems) for machine learning and more generally variational inequalities on which Gauthier has published several papers [Gidel et al. 2017, Gidel et al. 2018].

Vasilis Syrgkanis (MSR New England)

Vasilis is a Researcher at Microsoft Research, New England. His research lies at the intersection of theoretical computer science, machine learning and economics/econometrics. He received his Ph.D. in Computer Science from Cornell University and then spent two years as a postdoc researcher at Microsoft Research, NYC, as part of the Algorithmic Economics and the Machine Learning groups.
Vasilis has published numerous paper on algorithmic game theory including a best paper award at NeurIPS in 2015 for a paper providing fast convergence guarantees of regularized learning in games [Syrgkanis et al. 2015].

Éva Tardos (Cornell)

Éva Tardos is a Jacob Gould Schurman Professor of Computer Science at Cornell University, and she was Computer Science department chair from 2006 to 2010. She received her BA and PhD from Eötvös University in Budapest. She joined the faculty at Cornell in 1989. Tardos’s research interest is algorithms and algorithmic game theory. She is most known for her work on network-flow algorithms and quantifying the efficiency of selfish routing. She has been elected to the National Academy of Engineering, the National Academy of Sciences, the American Academy of Arts and Sciences, and is an external member of the Hungarian Academy of Sciences.
Éva is a leading figure in the algorithmic game theory field. She is the recipient of a number of fellowships and awards including the Packard Fellowship, the Gödel Prize, Dantzig Prize, Fulkerson Prize, ETACS prize, and the IEEE Technical Achievement Award. She is editor editor-in-Chief of the Journal of the ACM, has been editor-in-Chief of SIAM Journal of Computing, and editor of several other journals including Combinatorica; she served as problem committee member for many conferences, and was program committee chair for the ACM-SIAM Symposium on Discrete Algorithms (1996), as well as FOCS’05, and EC’13.

Léon Bottou (FAIR)

Léon Bottou received the Diplôme d’Ingénieur de l’École Polytechnique (X84) in 1987, the Magistère de Mathématiques Fondamentales et Appliquées et d’Informatique from École Normale Supérieure in 1988, and a Ph.D. in Computer Science from Université de Paris-Sud in 1991. His research career took him to AT&T Bell Laboratories, AT&T Labs Research, NEC Labs America and Microsoft. He joined Facebook AI Research in 2015.
The long term goal of Léon’s research is to understand how to build human-level intelligence. Although reaching this goal requires conceptual advances that cannot be anticipated at this point, it certainly entails clarifying how to learn and how to reason. Leon Bottou best known contributions are his work on “deep” neural networks in the 90s, his work on large scale learning in the 00’s, and possibly his more recent work on causal inference in learning systems. Leon has also recently worked on the geometry of probability measures applied to Generative Adversarial Networks.