DeepAGREL: Biologically plausible deep learning via direct reinforcement

Isabella Pozzi, Sander M. Bohte, Pieter R. Roelfsema

Keywords: biologically plausible deep learning, reinforcement learning, feedback gating, image claassification

TL;DR: We show how deep learning can be implemented in the brain using direct reinforcement learning just as well as error-backprop for hard tasks, with a surprisingly small penalty to the speed of convergence

Abstract:

While much recent work has focused on biologically plausible variants of error-backpropagation, learning in the brain seems to mostly adhere to a reinforcement learning paradigm; biologically plausible neural reinforcement learning frameworks, however, were limited to shallow networks learning from compact and abstract sensory representations. Here, we show that it is possible to generalize such approaches to deep networks with an arbitrary number of layers.
We demonstrate the learning scheme - DeepAGREL - on classical and hard image-classification benchmarks requiring deep networks, namely MNIST, CIFAR10, and CIFAR100, cast as direct reward tasks, both for deep fully connected, convolutional and locally connected architectures. We show that for these tasks, DeepAGREL achieves an accuracy that is equal to supervised error-backpropagation, and the trial-and-error nature of such learning imposes only a very limited cost in terms of training time. Thus, our results provide new insights into how deep learning may be implemented in the brain.