Censored Sampling for Topology Design: Guiding Diffusion with Human Preferences

2025-08-03

AI Summary

Existing diffusion models for topology optimization rely on surrogate constraints, limiting their ability to detect subtle physical defects—such as floating components and boundary discontinuities—that violate manufacturability or mechanical integrity. Method: We propose a human-in-the-loop diffusion framework: (i) a lightweight reward classifier is trained on sparse binary human preference feedback to accurately identify floating material and boundary violations; (ii) during sampling, gradient-based modulation of the reverse diffusion trajectory enables real-time design correction without retraining the generator. Contribution/Results: Our approach circumvents the limitations of surrogate modeling, significantly suppressing failure modes. It generates physically plausible, manufacturable structures across diverse scenarios. Crucially, it is the first method to efficiently integrate human prior knowledge directly into the diffusion sampling process—enhancing both generation quality and engineering practicality.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search
• Humans and AI: Learning Human Values and Preferences

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Fairness and ethical considerations in crowd work and in human-in-the-loop AI systems

Abstract

Recent advances in denoising diffusion models have enabled rapid generation of optimized structures for topology optimization. However, these models often rely on surrogate predictors to enforce physical constraints, which may fail to capture subtle yet critical design flaws such as floating components or boundary discontinuities that are obvious to human experts. In this work, we propose a novel human-in-the-loop diffusion framework that steers the generative process using a lightweight reward model trained on minimal human feedback. Inspired by preference alignment techniques in generative modeling, our method learns to suppress unrealistic outputs by modulating the reverse diffusion trajectory using gradients of human-aligned rewards. Specifically, we collect binary human evaluations of generated topologies and train classifiers to detect floating material and boundary violations. These reward models are then integrated into the sampling loop of a pre-trained diffusion generator, guiding it to produce designs that are not only structurally performant but also physically plausible and manufacturable. Our approach is modular and requires no retraining of the diffusion model. Preliminary results show substantial reductions in failure modes and improved design realism across diverse test conditions. This work bridges the gap between automated design generation and expert judgment, offering a scalable solution to trustworthy generative design.

Problem

• Guiding diffusion models with human preferences for topology design
• Detecting and suppressing unrealistic outputs in generative design
• Bridging automated design generation and expert judgment

Innovation

• Human-in-the-loop diffusion framework
• Lightweight reward model from human feedback
• Modular integration without retraining diffusion model