Title:Dequantified Diffusion Schrödinger Bridge for Density Ratio Estimation

Abstract:Density ratio estimation is fundamental to tasks involving $f$-divergences, yet existing methods often fail under significantly different distributions or inadequately overlap supports, suffering from the \textit{density-chasm} and the \textit{support-chasm} problems. Additionally, prior approaches yield divergent time scores near boundaries, leading to instability. We propose $\text{D}^3\text{RE}$, a unified framework for robust and efficient density ratio estimation. It introduces the Dequantified Diffusion-Bridge Interpolant (DDBI), which expands support coverage and stabilizes time scores via diffusion bridges and Gaussian dequantization. Building on DDBI, the Dequantified Schrödinger-Bridge Interpolant (DSBI) incorporates optimal transport to solve the Schrödinger bridge problem, enhancing accuracy and efficiency. Our method offers uniform approximation and bounded time scores in theory, and outperforms baselines empirically in mutual information and density estimation tasks.