Offline Contextual Bandits with Overparameterized Models

David Brandfonbrener · William Whitney · Rajesh Ranganath · Joan Bruna

Keywords: [ Non-convex Optimization ] [ Reinforcement Learning and Planning ] [ Stochastic Optimization ]

[ Abstract ]
[ Paper ]
[ Visit Poster at Spot A1 in Virtual World ]
Tue 20 Jul 9 a.m. PDT — 11 a.m. PDT
Spotlight presentation: Deep Reinforcement Learning 2
Tue 20 Jul 5 a.m. PDT — 6 a.m. PDT


Recent results in supervised learning suggest that while overparameterized models have the capacity to overfit, they in fact generalize quite well. We ask whether the same phenomenon occurs for offline contextual bandits. Our results are mixed. Value-based algorithms benefit from the same generalization behavior as overparameterized supervised learning, but policy-based algorithms do not. We show that this discrepancy is due to the \emph{action-stability} of their objectives. An objective is action-stable if there exists a prediction (action-value vector or action distribution) which is optimal no matter which action is observed. While value-based objectives are action-stable, policy-based objectives are unstable. We formally prove upper bounds on the regret of overparameterized value-based learning and lower bounds on the regret for policy-based algorithms. In our experiments with large neural networks, this gap between action-stable value-based objectives and unstable policy-based objectives leads to significant performance differences.

Chat is not available.