Lifelong Hyper-Policy Optimization with Multiple Importance Sampling Regularization

TL;DR

This paper introduces a lifelong reinforcement learning method that learns a hyper-policy to adapt to evolving dynamics, utilizing importance sampling and variance-based regularization to improve performance and mitigate forgetting.

Contribution

It proposes a novel hyper-policy framework that dynamically adapts policy parameters over time, incorporating importance sampling and variance regularization for continual learning.

Findings

Outperforms state-of-the-art algorithms in water resource management.

Effective in trading environments with evolving dynamics.

Reduces catastrophic forgetting through combined past and future performance estimates.

Abstract

Learning in a lifelong setting, where the dynamics continually evolve, is a hard challenge for current reinforcement learning algorithms. Yet this would be a much needed feature for practical applications. In this paper, we propose an approach which learns a hyper-policy, whose input is time, that outputs the parameters of the policy to be queried at that time. This hyper-policy is trained to maximize the estimated future performance, efficiently reusing past data by means of importance sampling, at the cost of introducing a controlled bias. We combine the future performance estimate with the past performance to mitigate catastrophic forgetting. To avoid overfitting the collected data, we derive a differentiable variance bound that we embed as a penalization term. Finally, we empirically validate our approach, in comparison with state-of-the-art algorithms, on realistic environments, including water resource management and trading.