Understanding Gap-Dependent Regret for Optimism-Based Reinforcement Learning with Linear Function Approximation

Name: Understanding Gap-Dependent Regret for Optimism-Based Reinforcement Learning with Linear Function Approximation
Uploaded: 2026-10-08T11:29:55.128192+00:00
Duration: 333 s
Description: The study improves performance guarantees for LSVI-UCB++, an optimism-based reinforcement learning algorithm using linear function approximation. It establishes a gap-dependent regret bound of Ob(d²H⁴/∆min), which is tighter than previous results, and shows that the required sample complexity can be reduced by improving bounds on estimation errors.

The study improves performance guarantees for LSVI-UCB++, an optimism-based reinforcement learning algorithm using linear function approximation. It establishes a gap-dependent regret bound of Ob(d²H⁴/∆min), which is tighter than previous results, and shows that the required sample complexity can be reduced by improving bounds on estimation errors.

Open the full episode