Predictive coding with spiking neurons and feedforward gist signalling

Abstract

Predictive coding (PC) is an influential theory in neuroscience, which suggests the existence of a cortical architecture that is constantly generating and updating predictive representations of sensory inputs. Owing to its hierarchical and generative nature, PC has inspired many computational models of perception in the literature. However, the biological plausibility of existing models has not been sufficiently explored due to their use of artificial neural network features such as a non-linear, continuous, and clock-driven function approximator as basic unit of computation. Therefore, we have developed a spiking neural network for predictive coding (SNN-PC), in which neurons communicate using event-driven and asynchronous spikes. While adopting the hierarchical structure and Hebbian learning algorithms from previous PC neural network models, SNN-PC introduces two novel features: 1) a fast feedforward sweep from the input to higher areas, which generates a spatially reduced and abstract representation of input (i.e., a neural code for the gist of a scene) and provides a neurobiological alternative to an arbitrary choice of priors; and 2) a separation of positive and negative error-computing neurons, which counters the biological implausibility of a bi-directional error neuron with a very high basal firing rate. After training with the MNIST handwritten digit dataset, SNN-PC developed hierarchical internal representations and was able to reconstruct samples it had not seen during training. SNN-PC suggests biologically plausible mechanisms by which the brain may perform perceptual inference and learning in an unsupervised manner. In addition, it may be used in neuromorphic applications that can utilize its energy-efficient, event-driven, local learning, and parallel information processing nature. Author summaryHow does the brain seamlessly perceive the world, in the midst of chaotic sensory barrage? Rather than passively relaying information that sensory organs pick up from the external world along the cortical hierarchy for a series of feature extractions, it actively gathers statistical regularities from sensory inputs to track causal relationships between physical properties of external objects and the body. In other words, the brains perceptual apparatus is constantly trying to make sense of the incoming streams of sensory input and represent the subjects current situation by building and maintaining internal models of the world and body. While this constructivist theme in understanding perception has been pervasive across multiple disciplines from philosophy to psychology to computer science, a comprehensive theory of brain function called predictive coding aims at unifying neural implementations of perception. In this study, we present a biologically plausible neural network for predictive coding that uses spiking neurons, Hebbian learning, and a feedforward visual pathway to perform perceptual inference and learning on images. Not only does the model show that predictive coding is well behaved under the biological constraint of spiking neurons, but it also provides deep learning and neuromorphic communities with novel paradigms of learning and computational architectures inspired by the natures most intelligent system, the brain.