Reproduce “Encoding Memory in World Models with Test-Time Model-Based 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

Maxime Burchi, Fayaz Ali Dharejo, Radu Timofte

TT-MBRL adapts a world model and policy to each evaluation episode through gradient updates using episode-specific data mixed with training data. On Memory Maze 15×15, Recall to Imagine with TT-MBRL scores 54.1 ± 3.5, versus 42.0 ± 5.8 without it, after 400 million environment steps. The score remains below the human baseline of 67.7.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.