• What is the outcome of multiagent learning in games? In this talk, we argue that convergence can be slippery.
• Joint work with David Gondek, Keith Hall, Amir Jafari, Michael Littman, Casey Marks, John Wicks, and Martin Zinkevich
  
• Presented at the Twentieth National Conference on Artificial Intelligence, July 2005.
• [ps] [pdf] [abstract]

Multiagent Value Iteration in Markov Games

• What is the outcome of multiagent learning in games? In this talk, we argue that convergence can be slippery.
• Joint work with Martin Zinkevich and Michael Littman.
• Presented at the The Stony Brook Game Theory Festival: Workshop on Game Theory and Computer Science, SUNY Stony Brook, July 2005.
• [ps] [pdf] [abstract]

Game-Theoretic Learning: Regret Minimization vs. Utility Maximization

• In this talk, we establish a fundamental connection between no-regret learning algorithms and game-theoretic equilibria.
• Joint work with David Gondek, Amir Jafari, and Casey Marks.
• Presented at University of California at Berkeley, Department of Computer Science, April 2005.
• Presented at University of Pennsylvania, Department of Computer Science, November 2004.
• Presented at Brown University, Wayland Collegium Interdisciplinary Faculty Seminar on Decision-Making and Rationality, October 2004.
• [ps] [pdf] [abstract]

• This talk describes an architecture developed for autonomous trading in online simultaneous auctions, inspired by TAC Travel.
• Joint work with Justin Boyan.
• Presented at New York University, Information Systems Research Seminar, December 2005.
• Presented at University of Michigan, STIET Seminar, September 2004.
• [ps] [pdf] [abstract]

Game-Theoretic Learning

• This tutorial reviews one of the fundamental connections between machine learning and game theory.
• Joint work with David Gondek, Amir Jafari, and Casey Marks.
• Presented at the International Conference on Machine Learning, Tutorial Program, July 2004. [abstract]
• Part I: Basic Game Theory [ps] [pdf]
• Part II: Regret Variations [ps] [pdf]
• Part III: Regret Matching Algorithms
• Part IV: Machine Learning Applications
• References [ps] [pdf]

• In this talk, we define a general class of no-regret learning algorithms, called no-Φ-regret learning algorithms, and a a class of game-theoretic equilibria, called Φ-equilibria, we show that for all φ ε Φ, the empirical distribution of play of no-φ-regret algorithms converges to the set of Φ-equilibria.
• [ps] [pdf]
• Presented at Caltech, October 2003. [abstract]
• Presented at COLT, August 2003. [abstract]
• Joint work with Amir Jafari.

Reinforcement Learning in Stochastic Games

• This talk advocates proposes an algorithm for computing correlated equilibrium policies in Markov games via no-regret learning.
• Presented at City University of New York, The Graduate Center, Seminar on Logic and Games, February 2003.
• Joint work with Keith Hall and David Gondek.
• [ps] [pdf] [abstract]
  

• This talk describes the case for learning correlated, rather than Nash, equilibrium policies in Markov games.
• Presented at NIPS Workshop on Multiagent Learning, December 2002.
• Joint work with Keith Hall.
• [ps] [pdf] [abstract]

Autonomous Bidding in the Trading Agent Competition

• This talk is concerned with the design of bidding policies for autonomous bidding agents in simultaneous auctions.
• Presented at Federal Communications Commission, Wireless Telecommunications Bureau, May 2002.
• Joint work with Justin Boyan.
• [ps] [pdf] [abstract]
  

Correlated and No-Regret Q-Learning

• This talk advocates proposes an algorithm for computing correlated equilibrium policies in Markov games via no-regret learning.
• Presented at Brown University, Workshop in Economic Theory, March 2002.
• Joint work with Keith Hall and David Gondek.
• [ps] [pdf] [abstract]
  

• This work is concerned with an algorithm called correlated-Q learning that converges to equilibrium policies in general-sum Markov games.
• Presented at DIMACS Workshop on Computational Issues in Game Theory and Mechanism Design, November 2001.
• Joint work with Keith Hall.
• [ps] [pdf] [abstract]

Bidding Algorithms for Simultaneous Auctions

• This talk is concerned with computational problems that arise in the design of bidding agents for simultaneous auctions.
• Presented at University of Maryland, Baltimore County, October 2001.
• Presented at IJCAI-2001 Workshop on Economic Agents, Models, and Mechanisms, August 2001.
• Joint work with Justin Boyan.
• [ps] [pdf] [abstract]
  

• This talk introduces RoxyBot, one of the top-scoring agents in the First International Trading Agent Competition.
• Presented at North Carolina State University, E-Commerce Seminar Series, November 2000.
• Presented at Brown University, Workshop in Economic Theory, November 2000.
• Presented at ACM Conference on Electronic Commerce, Trading Agent Competition Panel, October 2000.
• Joint work with Justin Boyan.
• [ps] [pdf] [abstract]
  

On the Connection between On-Line Learning and Games

• An empirical study of the relationship between game-theoretic solution concepts and on-line learning algorithms.
• Presented at Games 2000: The First Meeting of the Game Theory Society, Bilbao, Spain, July 2000.
• Presented at Workshop on Beyond MDPs: Representations and Algorithms, Uncertainty in Artificial Intelligence, Palo Alto, CA, June 2000.
• Joint work with Amir Jafari, Gunes Ercal, and David Gondek.
• [ps] [pdf] [abstract]
  

• This research concerns the impact of automated seller pricing on electronic markets. Extended Abstract.
• Presented at the Sixteenth International Joint Conference on Artificial Intelligence, Stockholm, Sweden, August 1999.
• Joint work with Jeff Kephart.
• [ps] [pdf] [abstract]
  

Shopbot Economics

• This research is concerned with the potential impact of automated buyer search on electronic markets.
• Presented at Workshop on Decision-Theoretic and Game-Theoretic Agents, University College London, July 1999.
• Joint work with Jeff Kephart.
• [ps] [pdf] [abstract]
  

Learning to Play Network Games

• This work identifies properties of learning algorithms which give rise to equilibrium outcomes in network games.
• New York University Dissertation Defense & Job Talk, May 1999.
• [ps] [pdf] [abstract]
  

• This research includes both negative and positive results on learning in the Santa Fe Bar Problem.
• Presented at the The Stony Brook Game Theory Festival: Workshop on Interactive Dynamics and Learning, SUNY Stony Brook, July 1998.
• Joint work with Bud Mishra and Rohit Parikh.
• [ps] [pdf] [abstract]
  

Learning in Networks Contexts: Experimental Results from Simulations

• This talk describes the results of simulations of a suite of learning algorithms in network contexts.
• Presented at the Eighth Meeting of the International Society of Dynamic Games, Maastrict, The Netherlands, July 1998.
• Presented at the Fourth INFORMS Telecommunications Conference, Boca Raton, Florida, March 1998.
• Joint work with Scott Shenker and Eric Friedman.
• [ps] [pdf] [abstract]
  

Learning to Play Network Games

• This presentation provides an introduction to the connection between learning, game theory, and networks.
• DARPA Graduate Student Workshop, Washington, DC, July 1998.
• [ps] [pdf] [abstract]