An Information-Theoretic Analysis of Out-of-Distribution Generalization in Meta-Learning with Applications to Meta-RL
This addresses generalization challenges in meta-learning, particularly for meta-reinforcement learning, but appears incremental as it builds on existing frameworks with theoretical analysis.
The paper tackles out-of-distribution generalization in meta-learning by analyzing it from an information-theoretic perspective, focusing on environment mismatch and broad-to-narrow training scenarios, and establishes generalization bounds for meta-reinforcement learning.
In this work, we study out-of-distribution generalization in meta-learning from an information-theoretic perspective. We focus on two scenarios: (i) when the testing environment mismatches the training environment, and (ii) when the training environment is broader than the testing environment. The first corresponds to the standard distribution mismatch setting, while the second reflects a broad-to-narrow training scenario. We further formalize the generalization problem in meta-reinforcement learning and establish corresponding generalization bounds. Finally, we analyze the generalization performance of a gradient-based meta-reinforcement learning algorithm.