Faster Convergence in Deep-Predictive-Coding Networks to Learn Deeper Representations

TL;DR

This paper introduces an accelerated inference method for deep-predictive-coding networks, enabling deeper architectures with improved feature representations and unsupervised object recognition performance.

Contribution

It provides a theoretical analysis of the inference bottleneck in DPCNs and proposes an accelerated inference strategy that enhances convergence and allows deeper, more effective networks.

Findings

Faster convergence guarantees for the new inference method.

Deep convolutional DPCNs outperform previous models in unsupervised object recognition.

Enhanced feature representations comparable to supervised convolutional networks.

Abstract

Deep-predictive-coding networks (DPCNs) are hierarchical, generative models. They rely on feed-forward and feed-back connections to modulate latent feature representations of stimuli in a dynamic and context-sensitive manner. A crucial element of DPCNs is a forward-backward inference procedure to uncover sparse, invariant features. However, this inference is a major computational bottleneck. It severely limits the network depth due to learning stagnation. Here, we prove why this bottleneck occurs. We then propose a new forward-inference strategy based on accelerated proximal gradients. This strategy has faster theoretical convergence guarantees than the one used for DPCNs. It overcomes learning stagnation. We also demonstrate that it permits constructing deep and wide predictive-coding networks. Such convolutional networks implement receptive fields that capture well the entire classes of objects on which the networks are trained. This improves the feature representations compared with our lab's previous non-convolutional and convolutional DPCNs. It yields unsupervised object recognition that surpass convolutional autoencoders and are on par with convolutional networks trained in a supervised manner.