Accelerated Markov Chain Monte Carlo Algorithms on Discrete States

Name: Accelerated Markov Chain Monte Carlo Algorithms on Discrete States
Uploaded: 2026-09-24T04:31:07.937483+00:00
Duration: 566 s
Description: The episode discusses a paper titled "Accelerated Markov Chain Monte Carlo Algorithms on Discrete States." The hosts explain how authors applied Nesterov acceleration to the Metropolis-Hastings algorithm, revealing a hidden gradient structure. They highlight the log-Fisher method as a practical improvement that converges faster and maintains positive probabilities in discrete state spaces.

The episode discusses a paper titled "Accelerated Markov Chain Monte Carlo Algorithms on Discrete States." The hosts explain how authors applied Nesterov acceleration to the Metropolis-Hastings algorithm, revealing a hidden gradient structure. They highlight the log-Fisher method as a practical improvement that converges faster and maintains positive probabilities in discrete state spaces.

Open the full episode