Associative Compression Networks

This paper introduces Associative Compression Networks (ACNs), a new framework for variational autoencoding with neural networks. The system differs from existing variational autoencoders in that the prior distribution used to model each code is conditioned on a similar code from the dataset. In compression terms this equates to sequentially transmitting the data using an ordering determined by proximity in latent space. As the prior need only account for local, rather than global variations in the latent space, the coding cost is greatly reduced, leading to rich, informative codes, even when autoregressive decoders are used. Experimental results on MNIST, CIFAR-10, ImageNet and CelebA show that ACNs can yield improved dataset compression relative to order-agnostic generative models, with an upper bound of 73.9 nats per image on binarized MNIST. They also demonstrate that ACNs learn high-level features such as object class, writing style, pose and facial expression, which can be used to cluster and classify the data, as well as to generate diverse and convincing samples.