Rényi-Regularized Flow-Based Variational Inference for Multimodal Posteriors
October 28, 2026
A new regularization technique using Rényi divergence helps normalizing flows better approximate complex, multimodal Bayesian posteriors. The method encourages the approximate posterior to cover the prior support, addressing failures in traditional Tikhonov calibration in well-separated mode scenarios.
HOW THIS AFFECTS YOU
●
researcherThis improves the ability of flow-based models to capture multiple modes in Bayesian inference.