Reproduce “Adaptive importance sampling for control and inference” 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
Hilbert Johan Kappen, Hans Christian Ruiz
The authors introduce PICE, a cross-entropy-based method that learns parametrized feedback controllers from trajectories for path-integral control. Controllers guide later sampling, linking better importance sampling with control. In a noisy inverted-pendulum example, effective sample size stabilizes at about 80% after 1,000 iterations. A neural-state inference example reports 60% ESS with feedback versus about 29% open-loop; these demonstrations concern limited problems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.