Improving the evaluation of samplers on multi-modal targets
Keywords: sampling from unnormalized densities, multi-modal density, diffusion models, variational inference
TL;DR: Proposes a systematic framework to evaluate sampling methods for multi-modal distributions, emphasizing mode separation, high dimensionality, and mode importance.
Abstract: Addressing multi-modality constitutes one of the major challenges of sampling. In this reflection paper, we advocate for a more systematic evaluation of samplers towards two sources of difficulty that are mode separation and dimension. For this, we propose a synthetic experimental setting that we illustrate on a selection of samplers, focusing on the challenging criterion of recovery of the mode relative importance. These evaluations are crucial to diagnose the potential of samplers to handle multi-modality and therefore to drive progress in the field.
Submission Number: 37
Loading