Convolution by Evolution: Differentiable Pattern Producing Networks

TL;DR

This paper introduces differentiable pattern producing networks (DPPNs) that combine evolution and learning to efficiently compress autoencoder weights and discover convolutional structures, improving generalization in vision tasks.

Contribution

The paper presents a novel differentiable network framework that evolves topology and learns weights, enabling automatic discovery of convolutional architectures within fully connected networks.

Findings

DPPNs compress autoencoder weights from 157,684 to about 200 parameters.

DPPNs achieve comparable reconstruction accuracy with far fewer parameters.

DPPNs generalize better across datasets, such as from MNIST to Omniglot.

Abstract

In this work we introduce a differentiable version of the Compositional Pattern Producing Network, called the DPPN. Unlike a standard CPPN, the topology of a DPPN is evolved but the weights are learned. A Lamarckian algorithm, that combines evolution and learning, produces DPPNs to reconstruct an image. Our main result is that DPPNs can be evolved/trained to compress the weights of a denoising autoencoder from 157684 to roughly 200 parameters, while achieving a reconstruction accuracy comparable to a fully connected network with more than two orders of magnitude more parameters. The regularization ability of the DPPN allows it to rediscover (approximate) convolutional network architectures embedded within a fully connected architecture. Such convolutional architectures are the current state of the art for many computer vision applications, so it is satisfying that DPPNs are capable of discovering this structure rather than having to build it in by design. DPPNs exhibit better generalization when tested on the Omniglot dataset after being trained on MNIST, than directly encoded fully connected autoencoders. DPPNs are therefore a new framework for integrating learning and evolution.