[ICLR 2022] Denoising Likelihood Score Matching for Condition Score-Based Data Generation
We propose a new denoising likelihood score-matching (DLSM) loss to deal with the score mismatch issue we found in the existing conditional score-based data generation methods.