Thermostat-assisted continuously-tempered Hamiltonian Monte Carlo for Bayesian learning
This work addresses challenges in Bayesian learning for large-scale and multimodal data, offering an incremental improvement over existing sampling methods.
The paper tackles the problem of Bayesian learning on large datasets and multimodal distributions by proposing a new sampling method, thermostat-assisted continuously-tempered Hamiltonian Monte Carlo, which efficiently draws i.i.d. samples and adaptively neutralizes mini-batch noise, demonstrating performance gains over strong baselines on three real datasets.
We propose a new sampling method, the thermostat-assisted continuously-tempered Hamiltonian Monte Carlo, for Bayesian learning on large datasets and multimodal distributions. It simulates the Nosé-Hoover dynamics of a continuously-tempered Hamiltonian system built on the distribution of interest. A significant advantage of this method is that it is not only able to efficiently draw representative i.i.d. samples when the distribution contains multiple isolated modes, but capable of adaptively neutralising the noise arising from mini-batches and maintaining accurate sampling. While the properties of this method have been studied using synthetic distributions, experiments on three real datasets also demonstrated the gain of performance over several strong baselines with various types of neural networks plunged in.