新研究收紧扩散模型评分匹配差距

报道来源 [7]

1. arXiv cs.AI TIER_1 English(EN) · Mansi, Avinash Kori, Francesco Leofante · 2026-07-07 04:00

   使用概念图在T2I扩散模型中实现高效偏见缓解

   arXiv:2607.03397v1 Announce Type: new Abstract: Text-to-Image diffusion models often propagate harmful bias inherited from the training data. Existing bias mitigation techniques typically intervene only at the text encoder or provide inference-time guidance, often leading to gene…

2. arXiv stat.ML TIER_1 English(EN) · Benjamin Dupuis, Tyler Farghly, Maxime Haddouche, Alain Durmus, Umut Simsekli · 2026-07-07 04:00

   缩小扩散模型的得分匹配差距

   arXiv:2607.04442v1 Announce Type: new Abstract: Diffusion models (DMs) are a state-of-the-art generative method to approximately sample from an unknown distribution. Their training and evaluation primarily rely on an Evidence Lower Bound (ELBO), which relates the Kullback-Leibler…

3. arXiv stat.ML TIER_1 English(EN) · Binxu Wang, Jacob Zavatone-Veth, Cengiz Pehlevan · 2026-07-07 04:00

   A Random Matrix Theory Perspective on the Consistency of Diffusion Models

   arXiv:2602.02908v2 Announce Type: replace-cross Abstract: Diffusion models trained on different, non-overlapping subsets of a dataset often produce strikingly similar outputs when given the same noise seed. We trace this consistency to a simple linear effect: the shared Gaussian …

4. arXiv cs.CV TIER_1 English(EN) · Patrick Mu Haojie · 2026-07-07 04:00

   A Decomposable Probe for Few-Step Diffusion Models: Prompt, Latent, and Score Selectivity across Backbone Families and Distillation Paradigms

   arXiv:2607.03256v1 Announce Type: new Abstract: Few-step distilled diffusion students cut text-to-image inference from ~50 to 1-8 network evaluations, but the quality gap is usually summarised by a single FID/CLIP scalar that cannot say which axis of the conditioning response cha…

5. arXiv cs.CV TIER_1 English(EN) · Ruchit Rawal, Reza Shirkavand, Sayak Paul, Yuxin Wen, Heng Huang, Yizheng Chen, Tom Goldstein, Gowthami Somepalli · 2026-07-07 04:00

   Flash-BoN: Instant Drafts for Inference-Time Scaling in Diffusion Models

   arXiv:2607.04461v1 Announce Type: new Abstract: Inference-time scaling for text-to-image generation has progressed from simple Best-of-$N$ (BoN) sampling to guided search methods that verify and steer candidate trajectories at intermediate denoising steps. These approaches focus …

6. arXiv cs.CV TIER_1 English(EN) · Shunqi Mao, Wei Guo, Chaoyi Zhang, Jieting Long, Ke Xie, Weidong Cai · 2026-07-07 04:00

   Ctrl-Z Sampling: Scaling Diffusion Sampling with Controlled Random Zigzag Explorations

   arXiv:2506.20294v5 Announce Type: replace Abstract: Diffusion models generate conditional samples by progressively denoising Gaussian noise, yet the denoising trajectory can stall at visually plausible but low-quality outcomes with conditional misalignment or structural artifacts…

7. arXiv stat.ML TIER_1 English(EN) · Umut Simsekli · 2026-07-05 18:00

   缩小扩散模型的得分匹配差距

   Diffusion models (DMs) are a state-of-the-art generative method to approximately sample from an unknown distribution. Their training and evaluation primarily rely on an Evidence Lower Bound (ELBO), which relates the Kullback-Leibler (KL) divergence of model samples to the score m…