New method efficiently attributes training data for diffusion models

Researchers have developed a new method called TID (Training-data Influence via score Discrepancy) to efficiently attribute training data for diffusion models. This approach estimates influence without costly per-sample gradient computations or retraining. A distilled version, TIDE, further reduces computational cost by training a forward-only student model to reproduce the teacher's rankings, enabling attribution in milliseconds. AI

IMPACT This method could significantly speed up the process of understanding how specific training data points influence generated outputs in diffusion models.

RANK_REASON The cluster contains an academic paper detailing a new method for diffusion models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• ArtBench-10
• arXiv
• CIFAR-10
• Denoising Diffusion Probabilistic Models
• Ms Coco
• TIDE

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →