A Deep Reinforcement Learning Approach to Marginalized Importance Sampling with the Successor Representation
This work addresses off-policy evaluation for reinforcement learning practitioners, offering a more scalable method compared to prior approaches.
The paper tackled the challenge of applying marginalized importance sampling (MIS) to high-dimensional domains by linking it to the successor representation, enabling stable off-policy evaluation in Atari and MuJoCo environments.
Marginalized importance sampling (MIS), which measures the density ratio between the state-action occupancy of a target policy and that of a sampling distribution, is a promising approach for off-policy evaluation. However, current state-of-the-art MIS methods rely on complex optimization tricks and succeed mostly on simple toy problems. We bridge the gap between MIS and deep reinforcement learning by observing that the density ratio can be computed from the successor representation of the target policy. The successor representation can be trained through deep reinforcement learning methodology and decouples the reward optimization from the dynamics of the environment, making the resulting algorithm stable and applicable to high-dimensional domains. We evaluate the empirical performance of our approach on a variety of challenging Atari and MuJoCo environments.