Hugging Face Daily Papers·· 2026-10-19AI score25
SWIFT: self-guided weak-to-strong diffusion model training via target extrapolation
SWIFT: Self-guided Weak-to-strong dIFfusion models via Target extrapolation
AI briefing
SWIFT is a self-bootstrapping training method that extrapolates clean regression targets using a frozen weak denoiser, improving stronger diffusion models without adding inference overhead. On ImageNet-256, it surpasses the baseline best FID with about half the total training compute; on RAEv2, it achieves an FID of 1.15 with unguided inference and 0.99 with guidance.
Open original arxiv.org