Synaptic Plasticity Dynamics for Deep Continuous Local Learning (DECOLLE)

Abstract

A growing body of work underlines striking similarities between biological neural networks and recurrent, binary neural networks. A relatively smaller body of work, however, addresses the similarities between learning dynamics employed in deep artificial neural networks and synaptic plasticity in spiking neural networks. The challenge preventing this is largely caused by the discrepancy between the dynamical properties of synaptic plasticity and the requirements for gradient backpropagation. Learning algorithms that approximate gradient backpropagation using local error functions can overcome this challenge. Here, we introduce Deep Continuous Local Learning (DECOLLE), a spiking neural network equipped with local error functions for online learning with no memory overhead for computing gradients. DECOLLE is capable of learning deep spatio temporal representations from spikes relying solely on local information, making it compatible with neurobiology and neuromorphic hardware. Synaptic plasticity rules are derived systematically from user-defined cost functions and neural dynamics by leveraging existing autodifferentiation methods of machine learning frameworks. We benchmark our approach on the event-based neuromorphic dataset N-MNIST and DvsGesture, on which DECOLLE performs comparably to the state-of-the-art. DECOLLE networks provide continuously learning machines that are relevant to biology and supportive of event-based, low-power computer vision architectures matching the accuracies of conventional computers on tasks where temporal precision and speed are essential.

A solution to the learning dilemma for recurrent networks of spiking neurons

“…Several methods for approximating stochastic gradient descent in feedforward networks of spiking neurons have been proposed, see e.g., refs. [40][41][42][43][44] . These employ-like e-prop-a pseudogradient to overcome the non-differentiability of a spiking neuron, as proposed previously in refs.…”

“…45 , 46 . References 40 , 42 , 43 arrive at a synaptic plasticity rule for feedforward networks that consists—like e-prop—of the product of a learning signal and a derivative (eligibility trace) that describes the dependence of a spike of a neuron j on the weight of an afferent synapse W j i . But in a recurrent network, the spike output of j depends on W j i also indirectly, via loops in the network that allow that a spike of neuron j contributes to the firing of other neurons, which in turn affect the firing of the presynaptic neuron i .…”

Brain-Inspired Learning on Neuromorphic Substrates

“…Although ignoring H E seems like a drastic simplification, several studies have used it to construct biologically plausible online learning rules as local approximations of RTRL. Empirically, these rules perform well on a number of complex problems either without recurrent connections, such as in the case of SuperSpike [17], or by ignoring gradient flow through the recurrent synaptic connectivity as done in e-Prop [27], RFLO [92], and DECOLLE [19]. These findings suggest that explicit recurrence is either not necessary for many problems or that ignoring explicit recurrence in gradient computations does not create a major impediment for successful learning, even when such recurrent connections are present.…”

Reinforcement learning in artificial and biological systems

“…Thus, an important challenge in bridging neuroscience and machine learning is to understand how plasticity processes can utilize this evaluative feedback efficiently for learning. Interestingly, an increasing body of work demonstrates that approximate forms of gradient backpropagation compatible with biological neural networks naturally incorporate such feedback, and models trained with them achieve near state-of-the-art results on classical classification benchmarks [89][90][91] . Synaptic plasticity rules can be derived from gradient descent that lead to 'threefactor' rules, consistent with an error-modulated Hebbian learning (Fig.…”