reSGLD

2000년 도입 · 논문 3편에서 사용

reSGLD proposes to simulate a high-temperature particle for exploration and a low-temperature particle for exploitation and allows them to swap simultaneously. Moreover, a correction term is included to avoid biases.

출처: Non-convex Learning via Replica Exchange Stochastic Gradient MCMC

소개 논문: Non-convex Learning via Replica Exchange Stochastic Gradient MCMC