MIT CSAIL's PoEM Predicts RL Policies on New Rewards Without Training
MIT CSAIL researchers introduced PoEM, a method that predicts the policy RL would learn on a new reward without actually training, by combining existing reward models, and found this works even when reward functions are not linearly related.
2026-11-21 ~ 2026-11-21 · 2 related posts
- MIT CSAIL's PoEM predicts RL outcomes on new rewards without any training — kastnerkyle · 2026-11-21
- PoEM: predicting RL policy outcomes by composing models trained on other rewards — gowthami_s · 2026-11-21