Retro Contest

OpenAI Blog · 8 years ago

A contest is launching to evaluate how well reinforcement learning algorithms can apply knowledge from prior training to new tasks. The competition measures transfer learning performance across multiple environments to test generalization capability. Winners will be determined by their algorithm's ability to reuse learned patterns rather than starting from scratch on novel problems.