DNNs, Dataset Statistics, and Correlation Functions
For machine learning theorists, it offers a new perspective on why DNNs generalize, potentially bridging physics and ML.
The paper argues that dataset correlational structure is key to DNN success in image recognition, proposing that DNNs discover high-order correlation functions, which explains their generalization beyond standard statistical learning theory.
This paper argues that dataset structure is important in image recognition tasks (among other tasks). Specifically, we focus on the nature and genesis of correlational structure in the actual datasets upon which DNNs are trained. We argue that DNNs are implementing a widespread methodology in condensed matter physics and materials science that focuses on mesoscale correlation structures that live between fundamental atomic/molecular scales and continuum scales. Specifically, we argue that DNNs that are successful in image classification must be discovering high order correlation functions. It is well-known that DNNs successfully generalize in apparent contravention of standard statistical learning theory. We consider the implications of our discussion for this puzzle.