Dmitry Baranchuk

Aligning DLM generation with reference distributions via feature-space MMD significantly boosts performance on perplexity and reasoning tasks without retraining the entire model.

Severe generation degradation in few-step parallel diffusion is an objective mismatch rather than an architectural limitation: swapping cross-entropy for a joint-aware $\alpha$-loss yields coherent generation in as few as four steps.

Achieving an FID of 3.98 in just two steps, the Interval Denoiser redefines the efficiency of latent-free image generation.