Do ideas have shape? Plato's theory of forms as the continuous limit of artificial neural networks

Abstract

We show that ResNets converge, in the infinite depth limit, to a generalization of image registration algorithms. In this generalization, images are replaced by abstractions (ideas) living in high dimensional RKHS spaces, and material points are replaced by data points. Whereas computational anatomy aligns images via deformations of the material space, this generalization aligns ideas by via transformations of their RKHS. This identification of ResNets as idea registration algorithms has several remarkable consequences. The search for good architectures can be reduced to that of good kernels, and we show that the composition of idea registration blocks with reduced equivariant multi-channel kernels (introduced here) recovers and generalizes CNNs to arbitrary spaces and groups of transformations. Minimizers of L2 regularized ResNets satisfy a discrete least action principle implying the near preservation of the norm of weights and biases across layers. The parameters of trained ResNets can be identified as solutions of an autonomous Hamiltonian system defined by the activation function and the architecture of the ANN. Momenta variables provide a sparse representation of the parameters of a ResNet. The registration regularization strategy provides a provably robust alternative to Dropout for ANNs. Pointwise RKHS error estimates lead to deterministic error estimates for ANNs.

Files

Additional details