Lei Feng

Context inconsistency in stepwise group-based RL can severely bias advantage estimation, but a hierarchical grouping strategy can fix it without extra compute.

Overcome simplicity bias in RL agents with PA-MoE, a mixture-of-experts architecture that learns task phases directly from the RL objective, leading to better expert specialization.

Multi-expert systems can suffer from *worse* performance than single-expert systems due to an inherent underfitting problem that arises from the difficulty of identifying the correct expert to defer to.