Model

Disco103

A representative result in algorithm self-discovery that uses evolutionary search or gradient-based meta-learning to discover general parameter update formulas. It surpasses Proximal Policy Optimization (PPO) on multiple standard reinforcement learning benchmarks.

Timeline (last 90 days)

Sep 30

The self-discovered algorithm Disco103 has already surpassed the classic PPO and achieved SOTA performance on multiple benchmarks.

Unverified50%

All Facts (1)

Unverified

The self-discovered algorithm Disco103 has already surpassed the classic PPO and achieved SOTA performance on multiple benchmarks.

50%

Source Articles

Demystifying Self-Discovered RL: Is Learning History an Asset or a Burden?Sep 30