Noise Hypernetworks: Amortizing Test-Time Compute in Diffusion Models

The new paradigm of test-time scaling has yielded remarkable breakthroughs in Large Language Models (LLMs) (e. g. reasoning models) and in generative vision models, allowing models to allocate additional computation during inference to effectively tackle increasingly complex problems. Despite the improvements of this approach, an important limitation emerges: the substantial increase in computation time makes the process slow and impractical for many applications. Given the success of this paradigm and its growing usage, we seek to preserve its benefits while eschewing the inference overhead. In this work we propose one solution to the critical problem of integrating test-time scaling knowledge into a model during post-training. Specifically, we replace reward guided test-time noise optimization in diffusion models with a Noise Hypernetwork that modulates initial input noise. We propose a theoretically grounded framework for learning this reward-tilted distribution for distilled generators, through a tractable noise-space objective that maintains fidelity to the base model while optimizing for desired characteristics. We show that our approach recovers a substantial portion of the quality gains from explicit test-time optimization at a fraction of the computational cost.

Authors

• Luca Eyring
• Shyamgopal Karthik
• Alexey Dosovitskiy
• Nataniel Ruiz Boston University
• Zeynep Akata

Keywords

No keywords are indexed for this paper.