SWIFT: Alibaba and UMich paper cuts diffusion training cost via weak-denoiser target extrapolation
alibabagroup · hf · 2026-10-21
SWIFT, from University of Michigan, Alibaba Group, UC Berkeley and Nanjing University, is a self-guided training method that uses a frozen weak denoiser to improve a stronger diffusion model: for each noisy input, it extrapolates the clean regression target away from the weak denoiser's prediction, amplifying underfit components.
- The weak guide can come from an early-training checkpoint, requires no backpropagation during main-model training, and is discarded afterward
- The resulting model keeps the original sampler and supports conditional and unconditional generation without label dropout or an unconditional branch
- The mechanism is explained via spectral analysis of finite-time population gradient flow in an RKHS
- Reduces error and training cost with zero inference overhead
More from Multimodal
- EvaOmni: an evidence-grounded benchmark for audio-visual reasoning in omnimodal LLMs — VisionXLab · 2026-10-21
- WinARDet: compressed window autoregression stops box repetition loops in dense grounding — NankaiUniversity · 2026-10-21
- Qwen-Image 2.1 edit keeps producing plastic skin, negative prompts don't help — NorthAppropriate4458 · 2026-10-21
- AI-generated illustration visualizes Paz Soldán's sci-fi story 'Artificial' and its Ship of Theseus theme — dasein47 · 2026-10-21
- AI-generated Halloween video of a woman walking through skeleton crowds goes viral — Stephane_AI · 2026-10-21
- AI-generated fantasy short Lost In Woods stuns with cinematic forest fairy scene — Some-Dark-5802 · 2026-10-21