HMC with Normalizing Flows
This work addresses efficiency in sampling for computational physics, but it appears incremental as it adapts existing methods (Normalizing Flows) to a known bottleneck in HMC.
The authors tackled the problem of generating independent configurations in Hamiltonian Monte Carlo (HMC) by using Normalizing Flows as a trainable kernel to simplify dynamics, resulting in outperformance of traditional methods with scalability to large lattice volumes and minimal retraining effort.
We propose using Normalizing Flows as a trainable kernel within the molecular dynamics update of Hamiltonian Monte Carlo (HMC). By learning (invertible) transformations that simplify our dynamics, we can outperform traditional methods at generating independent configurations. We show that, using a carefully constructed network architecture, our approach can be easily scaled to large lattice volumes with minimal retraining effort. The source code for our implementation is publicly available online at https://github.com/nftqcd/fthmc.