Rapid mixing in positively weighted restricted Boltzmann machines

This addresses a computational bottleneck for researchers and practitioners in machine learning who rely on efficient sampling for training and inference in probabilistic models, though it is incremental as it builds on existing methods for ferromagnetic systems.

The paper tackled the problem of slow mixing times in Markov chain Monte Carlo samplers for restricted Boltzmann machines by proving polylogarithmic mixing time bounds for the alternating-scan sampler in positively weighted cases, achieving results up to critical thresholds.