Decomposing neural networks as mappings of correlation functions
Abstract
Understanding the functional principles of information processing in deep neural networks continues to be a challenge, in particular for networks with trained and thus non-random weights. To address this issue, we study the mapping between probability distributions implemented by a deep feed-forward network. We characterize this mapping as an iterated transformation of distributions, where the non-linearity in each layer transfers information between different orders of correlation functions. This allows us to identify essential statistics in the data, as well as different information representations that can be used by neural networks. Applied to an XOR task and to MNIST, we show that correlations up to second order predominantly capture the information processing in the internal layers, while the input layer also extracts higher-order correlations from the data. This analysis provides a quantitative and explainable perspective on classification.
Keywords
Cite
@article{arxiv.2202.04925,
title = {Decomposing neural networks as mappings of correlation functions},
author = {Kirsten Fischer and Alexandre René and Christian Keup and Moritz Layer and David Dahmen and Moritz Helias},
journal= {arXiv preprint arXiv:2202.04925},
year = {2023}
}
Comments
Published in Physical Review Research Changes with respect to the previous version: - Added results with CIFAR-10 - Added sections to the supplementary: - Derivation of an analogous result to the depth scale of untrained deep networks. - Expanded discussion applicability of the Gaussian assumption when variables are weakly correlated. - Clarified main text in some areas. - Fixed typos