On the Feature Learning in Diffusion Models
This work provides theoretical and empirical insights into the feature learning mechanisms of diffusion models, which is incremental for researchers in generative modeling and representation learning.
The paper tackles the problem of understanding feature learning dynamics in diffusion models compared to classification models, finding that diffusion models learn more balanced and comprehensive representations due to their denoising objective, while classification models prioritize specific patterns, with experiments on synthetic and real-world datasets validating these insights.
The predominant success of diffusion models in generative modeling has spurred significant interest in understanding their theoretical foundations. In this work, we propose a feature learning framework aimed at analyzing and comparing the training dynamics of diffusion models with those of traditional classification models. Our theoretical analysis demonstrates that diffusion models, due to the denoising objective, are encouraged to learn more balanced and comprehensive representations of the data. In contrast, neural networks with a similar architecture trained for classification tend to prioritize learning specific patterns in the data, often focusing on easy-to-learn components. To support these theoretical insights, we conduct several experiments on both synthetic and real-world datasets, which empirically validate our findings and highlight the distinct feature learning dynamics in diffusion models compared to classification.