PhD student, Korea Advanced Institute of Science & Technology
1 paper at NeurIPS 2025
We propose KLASS, a fast KL-guided sampling method for masked diffusion models that improves accuracy while cutting inference time by over 2x.
