Mechanistic Interpretability of Diffusion Models: Circuit-Level Analysis and Causal Validation

TL;DR

This paper provides a detailed circuit-level analysis of diffusion models, revealing how they process images differently for synthetic and real data, and identifies key attention mechanisms with causal validation.

Contribution

It introduces a systematic, quantitative approach to understanding diffusion models at the circuit level, including causal validation of specific computational pathways.

Findings

Real-world face processing circuits have higher computational complexity.

Eight distinct attention mechanisms with specialized roles were identified.

Targeted interventions cause significant performance degradation, confirming causal circuit functions.

Abstract

We present a quantitative circuit-level analysis of diffusion models, establishing computational pathways and mechanistic principles underlying image generation processes. Through systematic intervention experiments across 2,000 synthetic and 2,000 CelebA facial images, we discover fundamental algorithmic differences in how diffusion architectures process synthetic versus naturalistic data distributions. Our investigation reveals that real-world face processing requires circuits with measurably higher computational complexity (complexity ratio = 1.084 plus/minus 0.008, p < 0.001), exhibiting distinct attention specialization patterns with entropy divergence ranging from 0.015 to 0.166 across denoising timesteps. We identify eight functionally distinct attention mechanisms showing specialized computational roles: edge detection (entropy = 3.18 plus/minus 0.12), texture analysis (entropy = 4.16 plus/minus 0.08), and semantic understanding (entropy = 2.67 plus/minus 0.15). Intervention analysis demonstrates critical computational bottlenecks where targeted ablations produce 25.6% to 128.3% performance degradation, providing causal evidence for identified circuit functions. These findings establish quantitative foundations for algorithmic understanding and control of generative model behavior through mechanistic intervention strategies.