arXiv:2509.08933v2 Announce Type: replace Abstract: We study the problem of learning the optimal policy in a discounted, infinite-horizon reinforcement learning (RL) setting in the presence of adversarially corrupted rewards. To address this problem, we develop a novel robust variant of the \(Q\)-learning algorithm and analyze it under the challenging asynchronous sampling model with time-correlated data. Despite corruption, we prove that the finite-time guarantees of our approach match existing bounds, up to an additive term that scales with the fraction of corrupted samples. We also establis
This research is emerging as AI systems are increasingly deployed in real-world, potentially adversarial environments where data integrity cannot be guaranteed.
It demonstrates a significant step towards developing more resilient and reliable AI agents, crucial for deployment in sensitive or high-stakes applications.
The ability to develop robust reinforcement learning algorithms that can tolerate data corruption opens new pathways for more secure and dependable autonomous systems.
- · AI developers
- · Robotics
- · Critical infrastructure AI
- · Defence contractors
- · Adversarial actors
- · AI systems susceptible to data corruption
More secure and robust AI agent deployments across various sectors.
Reduced vulnerability of autonomous systems to intentional attacks or unintentional data glitches.
Accelerated adoption of AI in sectors requiring high integrity and resilience, potentially shifting competitive advantages.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG