Hugging Face Daily Papers·· 2026-10-19AI score33
SpecKCast: accelerating diffusion model inference with spectral Kalman prediction
SpecKCast: Spectral Kalman Forecasting for Diffusion Acceleration
AI briefing
SpecKCast is a training-free method that predicts diffusion model outputs in the frequency domain to skip redundant denoising steps: it applies a 2D DCT to the output, then uses a lightweight Kalman filter to track each coefficient's value and rate of change along band-time coordinates derived from the noise schedule, invoking the transformer only when prediction uncertainty exceeds a threshold.
Open original arxiv.org