Morning Edition · Saturday, October 17, 2026Published at 2:12 AM EDT · New York

Training-free VISTA method chooses which denoising steps receive Sequential Monte Carlo guidance under a fixed compute budget

The authors replace fixed sparse-guidance heuristics with a dynamic program that minimizes a bound on deviation from full guidance, and they test it on text, biological sequences and images.

Reward-guided discrete diffusion lets a model align its outputs with a downstream objective at inference time, without retraining. Sequential Monte Carlo (SMC) guidance is a common way to do this. Each guided step needs extra reward evaluat…