Paper

Robust Diffusion Models via Divergence-Induced Weighted Denoising

Unreadunread

AI summary

arXiv:2606. 22521v1 Announce Type: new Abstract: We show that replacing the standard MSE denoising loss in diffusion models with a nonlinear transformation induced by an f-divergence yields a simple robust training surrogate that empirically improves performance under data contamination, with small additional computational overhead.

Reading status

Priority

About this paper

Authors: Lei Li, Yuexiao Dong
Released: 2026-06-23 (1mo ago)
Topic: Related to current project
Relation to my work: Read next because Robust Diffusion Models via Divergence-Induced Weighted Denoising overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)", experiment "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone". Matching terms: under, line, control, model. Source: arxiv stat.ML (Machine Learning).
arXiv: 2606.22521

Ask Claude about this paper

Posts a comment with @claude so the runner spins up a Claude Code session scoped to Robust Diffusion Models via Divergence-Induced Weighted Denoising. Replies appear in the conversation below.

Comments

No comments yet.

Related records

No links yet.