Mechanistic Interpretability of Diffusion Models: Circuit-Level Analysis and Causal Validation
This work provides quantitative foundations for algorithmic understanding and control of generative models, addressing the problem of interpretability for researchers and practitioners in AI, though it is incremental in building on existing mechanistic interpretability methods.
The paper tackled the problem of understanding the computational pathways in diffusion models for image generation, discovering fundamental algorithmic differences in processing synthetic versus naturalistic data, with real-world face processing requiring circuits of higher complexity and specialized attention mechanisms, leading to performance degradations of 25.6% to 128.3% upon targeted ablations.
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.