Stagewise Reinforcement Learning and the Geometry of the Regret Landscape
Organizations: Timaeus · University of Oxford
Abstract
Singular learning theory characterizes Bayesian learning as an evolving tradeoff between accuracy and complexity, with transitions between qualitatively different solutions as sample size increases. We extend this theory to reinforcement learning, proving that the concentration of a generalized posterior over policies is governed by the local learning coefficient (LLC), an invariant of the geometry of the regret function. This theory predicts that deep reinforcement learning with SGD should proceed from simple policies with high regret to complex policies with low regret. We verify this prediction empirically in a gridworld environment exhibiting stagewise policy development: phase transitions over training manifest as "opposing staircases" where regret decreases sharply while the LLC increases.
Figures & tables
Appendix figures & tables7 assets
Supplementary material from the paper’s appendix.
Appendix
| Parameter | Value |
|---|---|
| Rollouts | |
| # parallel environments | 9600 |
| Rollout length | 64 |
| # environment steps per gradient step | 614.4k |
| Discount factor, | 0.97, 0.98, 0.99 |
| REINFORCE updates |
| Notation | Description |
|---|---|
| Initial point in | |
| Inverse temperature | |
| Number of training steps | |
| Noise variance | |
| SGLD chain length | |
| SGLD step size |
Reinforcement Learning
Reinforcement learning studies how agents learn sequential decisions from rewards and interaction, balancing exploration, performance, safety, and adaptation.
See all 1,847 papers