Learning without feedback: Fixed random learning signals allow for feedforward training of deep neural networks

TL;DR

This paper introduces DRTP, a feedforward training method for deep neural networks using fixed random signals, eliminating the need for backpropagation and enabling efficient training suitable for edge devices.

Contribution

The paper proposes the DRTP algorithm that uses fixed random projections of labels as targets, addressing backpropagation constraints and facilitating low-cost, layerwise training.

Findings

DRTP matches accuracy of backpropagation in certain tasks

Reduces memory and computational overhead for training

Enables training of deep networks on edge devices

Abstract

While the backpropagation of error algorithm enables deep neural network training, it implies (i) bidirectional synaptic weight transport and (ii) update locking until the forward and backward passes are completed. Not only do these constraints preclude biological plausibility, but they also hinder the development of low-cost adaptive smart sensors at the edge, as they severely constrain memory accesses and entail buffering overhead. In this work, we show that the one-hot-encoded labels provided in supervised classification problems, denoted as targets, can be viewed as a proxy for the error sign. Therefore, their fixed random projections enable a layerwise feedforward training of the hidden layers, thus solving the weight transport and update locking problems while relaxing the computational and memory requirements. Based on these observations, we propose the direct random target projection (DRTP) algorithm and demonstrate that it provides a tradeoff between accuracy and computational cost that is suitable for adaptive edge computing devices.