Timezone: »

 
Corruption Robust Offline Reinforcement Learning
Xuezhou Zhang · Yiding Chen · Jerry Zhu · Wen Sun

@
in Reinforcement Learning for Real Life »
We study the adversarial robustness in offline reinforcement learning. Given a batch dataset consisting of tuples $(s, a, r, s')$, an adversary is allowed to arbitrarily modify $\epsilon$ fraction of the tuples. From the corrupted dataset the learner aims to robustly identify a near-optimal policy. We first show that a worst-case $\Omega(d\epsilon)$ optimality gap is unavoidable in linear MDP of dimension $d$, even if the adversary only corrupts the reward element in a tuple. This contrasts with dimension-free results in robust supervised learning and best-known lower-bound in the online RL setting with corruption. Next, we propose a robust variant of the Least-Square Value Iteration (LSVI) algorithm utilizing robust supervised learning oracles, that achieves near-matching performances in cases both with and without full data coverage. The algorithm requires the knowledge of $\epsilon$ to design the pessimism bonus in the no-coverage case. Surprisingly, in this case, the knowledge of $\epsilon$ is necessary, as we show that being adaptive to unknown $\epsilon$ is impossible. This again contrasts with recent results on corruption-robust online RL and implies that robust offline RL is a strictly harder problem.

Author Information

Xuezhou Zhang (UW-Madison)
Yiding Chen (University of Wisconsin-Madison)
Jerry Zhu (University of Wisconsin-Madison)
Wen Sun (Cornell University)

More from the Same Authors