Low Regret

Low regret, a central theme in online learning and decision-making, focuses on minimizing the cumulative difference between an algorithm's performance and that of an optimal strategy chosen with hindsight. Current research emphasizes developing algorithms with provably low regret across various settings, including multi-armed bandits, reinforcement learning, and game theory, often employing techniques like optimistic exploration, primal-dual methods, and information-theoretic bounds. These advancements are crucial for improving the efficiency and robustness of decision-making systems in diverse applications, ranging from personalized recommendations and resource allocation to control systems and AI safety. The pursuit of low regret is driving innovation in algorithm design and theoretical analysis, leading to more efficient and reliable learning systems.

Papers

February 2, 2022

Non-Stationary Dueling Bandits
Patrick Kolpaczki, Viktor Bengs, Eyke Hüllermeier
Non Stationary Bandit Algorithm Dueling Bandit Low Regret

December 28, 2021

Efficient Performance Bounds for Primal-Dual Reinforcement Learning from Demonstrations
Angeliki Kamoutsi, Goran Banjac, John Lygeros
Inverse Reinforcement Learning Noisy Demonstration Policy Optimization Stochastic Convex Optimization Low Regret Performance Bound Primal Dual Reinforcement Learning

November 16, 2021

Learning Equilibria in Mean-Field Games: Introducing Mean-Field PSRO
Paul Muller, Mark Rowland, Romuald Elie, Georgios Piliouras, Julien Perolat, Mathieu Lauriere, Raphael Marinier, Olivier Pietquin, Karl Tuyls
Mean Field Multi Agent Learning Mean Field Game Low Regret Learning Equilibrium

Low Regret

Papers

Non-Stationary Dueling Bandits

Efficient Performance Bounds for Primal-Dual Reinforcement Learning from Demonstrations

Learning Equilibria in Mean-Field Games: Introducing Mean-Field PSRO