Density Ratio Estimation

Density ratio estimation (DRE) focuses on determining the ratio between two probability distributions, providing a powerful tool for comparing and contrasting data from different sources. Current research emphasizes improving the accuracy and robustness of DRE, particularly in high-dimensional spaces, using techniques like neural networks trained with various divergence-based loss functions (e.g., f-divergences, α-divergence) and exploring connections to other machine learning tasks such as classification and generative modeling. DRE's impact spans diverse fields, enabling improved synthetic data evaluation, robust uncertainty quantification in medical imaging, and enhanced performance in domain adaptation and off-policy evaluation.