Reproduce “Compression Forcing: Hierarchical History Tokenization Enables Flexible World Models” 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
Siqiao Huang, Partha Kaushik, Michael Chen, Hengkai Pan, Kaiwen Geng, Omar Chehab, Fernando Moreno-Pino, Max Simchowitz
Compression Forcing organizes video history into ordered tokens, prioritizing changes before static detail. Dynamics-Aware Tokenization trains short prefixes for motion prediction and longer prefixes for reconstruction; Hierarchical History Guidance combines predictions across prefix lengths. On Kinetics-600, it reaches 18.83 dB PSNR versus 18.60 for ElasticTok. In simulated drawer evaluation, predicted success is 64%, below the real 81.5%.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.