On the relationship between predictive coding and backpropagation

TL;DR

This paper explores the mathematical relationship between predictive coding and backpropagation, suggesting predictive coding as a biologically plausible alternative for training neural networks, and discusses its implications for biological learning models.

Contribution

It provides a detailed mathematical analysis linking predictive coding to backpropagation and introduces Torch2PC, a tool for implementing predictive coding in PyTorch.

Findings

Predictive coding can approximate backpropagation in neural network training.

The work offers insights into biological plausibility of neural network learning algorithms.

A new software tool, Torch2PC, facilitates predictive coding implementations.

Abstract

Artificial neural networks are often interpreted as abstract models of biological neuronal networks, but they are typically trained using the biologically unrealistic backpropagation algorithm and its variants. Predictive coding has been proposed as a potentially more biologically realistic alternative to backpropagation for training neural networks. This manuscript reviews and extends recent work on the mathematical relationship between predictive coding and backpropagation for training feedforward artificial neural networks on supervised learning tasks. Implications of these results for the interpretation of predictive coding and deep neural networks as models of biological learning are discussed along with a repository of functions, Torch2PC, for performing predictive coding with PyTorch neural network models.