Reproduce “Strategic Attentive Writer for Learning Macro-Actions” 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
Alexander, Vezhnevets, Volodymyr Mnih, John Agapiou, Simon Osindero, Alex Graves, Oriol Vinyals, Koray Kavukcuoglu
STRAW learns multi-step action plans and when to revise them from task rewards. It represents future actions as a plan and learns when to re-plan, allowing commitments to vary with the situation. In Atari tests, STRAWe scored 8,108 on Frostbite versus 1,281 for the LSTM baseline; STRAW or STRAWe led on six of eight games, with Hero and Breakout exceptions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.