Interacting Contour Stochastic Gradient Langevin Dynamics
This work addresses computational bottlenecks in Bayesian inference for large datasets, though it appears incremental as an extension of existing contour stochastic gradient Langevin dynamics.
The authors tackled the problem of inefficient posterior sampling in large-scale uncertainty estimation by proposing an interacting contour stochastic gradient Langevin dynamics (ICSGLD) sampler, which showed theoretical efficiency gains over single-chain methods and demonstrated great potential in numerical comparisons with benchmark algorithms.
We propose an interacting contour stochastic gradient Langevin dynamics (ICSGLD) sampler, an embarrassingly parallel multiple-chain contour stochastic gradient Langevin dynamics (CSGLD) sampler with efficient interactions. We show that ICSGLD can be theoretically more efficient than a single-chain CSGLD with an equivalent computational budget. We also present a novel random-field function, which facilitates the estimation of self-adapting parameters in big data and obtains free mode explorations. Empirically, we compare the proposed algorithm with popular benchmark methods for posterior sampling. The numerical results show a great potential of ICSGLD for large-scale uncertainty estimation tasks.