DDNO: Discrete Diffusion Noise Optimization

MCML Authors

Luca Eyring

→ Group Zeynep Akata
 Interpretable and Reliable Machine Learning

Vincent Pauline

→ Group Stefan Bauer
 Algorithmic Machine Learning & Explainable AI

Stefan Bauer

Prof. Dr.

Core PI

Algorithmic Machine Learning & Explainable AI

Zeynep Akata

Prof. Dr.

Core PI

Interpretable and Reliable Machine Learning

Abstract

Aligning discrete diffusion models with downstream rewards remains challenging: step-wise guidance is myopic and degrades sample quality, while fine-tuning is expensive and task-specific. We introduce Discrete Diffusion Noise Optimization (DDNO), a training-free method that instead optimizes the initial discrete noise to maximize terminal rewards while keeping the generator frozen. DDNO parameter- izes the noise distribution with continuous logits and propagates gradients through the reverse process via a straight-through surrogate combined with soft mixing, enabling stable optimization over long denoising trajectories. On compositional text-to-image synthesis and controllable text generation, DDNO consistently out- performs inference-time baselines like guidance and Best-of-N while exhibiting favorable scaling. This positions DDNO as a promising axis for test-time scaling in discrete generative models, complementing advances in continuous diffusion.

inproceedings EPB+26

---

ReALM-GEN @ICLR 2026

Workshop on Real‑World Constrained and Preference‑Aligned Flow‑ and Diffusion‑Based Models at the 14th International Conference on Learning Representations. Rio de Janeiro, Brazil, Apr 23-27, 2026. To be published. Preprint available.

---

Authors

L. Eyring • V. Pauline • S. Bauer • A. Dosovitskiy • Z. Akata

---

Links

URL

---

Research Areas

 A1 | Statistical Foundations & Explainability

 B1 | Computer Vision