Faster Convergence in Deep-Predictive-Coding Networks to Learn Deeper Representations
This addresses a computational bottleneck for researchers using DPCNs, though it appears incremental as it builds on the lab's previous work.
The paper tackles the computational bottleneck in deep-predictive-coding networks (DPCNs) that limits network depth due to learning stagnation, and proposes a new forward-inference strategy based on accelerated proximal gradients, which enables constructing deep and wide networks that yield unsupervised object recognition on par with supervised convolutional networks.
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.