Neural Networks on Groups
share
Recent work on neural networks has shown that allowing them to build internal representations of data not restricted to R^n can provide significant improvements in performance. The success of Graph Neural Networks, Convolutional Kernel Networks, and Fourier Neural Networks among other methods have demonstrated the clear value of applying abstract mathematics to the design of neural networks. The theory of neural networks has not kept up however, and the relevant theoretical results (when they exist at all) have been proven on a case-by-case basis without a general theory. The process of deriving new theoretical backing for each new type of network has become a bottleneck to understanding and validating new approaches. In this paper we extend the concept of neural networks to general groups and prove that neural networks with a single hidden layer and a bounded non-constant activation function can approximate any L^p function defined over a locally compact Abelian group. This framework and universal approximation theorem encompass all of the aforementioned contexts. We also derive important corollaries and extensions with minor modification, including the case for approximating continuous functions on a compact subset, neural networks with ReLU activation functions on a linearly bi-ordered group, and neural networks with affine transformations on a vector space. Our work also obtains as special cases the recent theorems of Qi et al. [2017], Sennai et al. [2019], Keriven and Peyre [2019], and Maron et al. [2019].
READ FULL TEXT