Reproduce “Learning Shared Representations in Multi-task Reinforcement Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Diana Borsa, Thore Graepel, John Shawe-Taylor
The authors develop multi-task reinforcement-learning methods that learn shared state–action features while fitting a separate value function for each task, with sparse task-specific features to capture differences. In four-room navigation, joint methods often achieved better normalized reward than single-task baselines with 500 samples per task; transfer helped on new goals, but these results do not establish scalability beyond this setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.