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.
Abstract (translated)
我们对扩散模型进行了定量的电路级分析，建立了图像生成过程背后的计算路径和机制原理。通过在2000张合成图像和2000张CelebA面部图像上进行系统干预实验，我们发现了扩散架构处理合成数据与自然数据分布时的基本算法差异。我们的研究发现，对现实世界人脸的处理需要具有可测量更高计算复杂度（复杂性比率 = 1.084 ± 0.008, p < 0.001）的电路，并表现出注意力专门化模式，去噪步骤间的熵散度范围为0.015至0.166。我们识别出了八种功能上不同的注意机制，显示出特定的计算角色：边缘检测（熵 = 3.18 ± 0.12），纹理分析（熵 = 4.16 ± 0.08）和语义理解（熵 = 2.67 ± 0.15）。干预分析展示了关键的计算瓶颈，目标消融会导致性能下降25.6%到128.3%，为所识别电路功能提供了因果证据。这些发现通过机制性干预策略建立了算法理解和控制生成模型行为的量化基础。
URL
https://arxiv.org/abs/2506.17237
