Reproduce “Dynamics Aware Past-Future Information Bottleneck” 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
Wooyoung Chung, Stas Tiomkin
The authors propose predictive compression that preserves a system’s block-Hankel past-to-future map during optimization, so the resulting predictor has a state-space realization. In simulations, its realized information curve overlaps the structured curve and stays close to the unconstrained optimum. At β = 2.06, trajectory RMS was 1.59, versus 8.33 for standard PF-IB, in a linear two-mass system.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.