Reproduce “Metacontrol for Adaptive Imagination-Based Optimization” 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
Jessica B. Hamrick, Andrew J. Ballard, Razvan Pascanu, Oriol Vinyals, Nicolas Heess, Peter W. Battaglia
The metacontroller allocates internal computation by proposing controls, evaluating them with predictive experts, and deciding whether to ponder again or act. In a one-shot simulated spaceship task, its total cost—task loss plus assigned penalties for expert calls—was 11% and 15% lower than the best fixed-iteration agents using experts; the paper reports these reductions inconsistently as median and average.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.