Abstract:In recent years the field of neuromorphic low-power systems that consume orders of magnitude less power gained significant momentum. However, their wider use is still hindered by the lack of algorithms that can harness the strengths of such architectures. While neuromorphic adaptations of representation learning algorithms are now emerging, efficient processing of temporal sequences or variable length-inputs remain difficult. Recurrent neural networks (RNN) are widely used in machine learning to solve a variety of sequence learning tasks. In this work we present a train-and-constrain methodology that enables the mapping of machine learned (Elman) RNNs on a substrate of spiking neurons, while being compatible with the capabilities of current and near-future neuromorphic systems. This "train-and-constrain" method consists of first training RNNs using backpropagation through time, then discretizing the weights and finally converting them to spiking RNNs by matching the responses of artificial neurons with those of the spiking neurons. We demonstrate our approach by mapping a natural language processing task (question classification), where we demonstrate the entire mapping process of the recurrent layer of the network on IBM's Neurosynaptic System "TrueNorth", a spike-based digital neuromorphic hardware architecture. TrueNorth imposes specific constraints on connectivity, neural and synaptic parameters. To satisfy these constraints, it was necessary to discretize the synaptic weights and neural activities to 16 levels, and to limit fan-in to 64 inputs. We find that short synaptic delays are sufficient to implement the dynamical (temporal) aspect of the RNN in the question classification task. The hardware-constrained model achieved 74% accuracy in question classification while using less than 0.025% of the cores on one TrueNorth chip, resulting in an estimated power consumption of ~17 uW.

Improving Spiking Neural Network Accuracy Using Time-based Neurons

Energy-Efficient High-Accuracy Spiking Neural Network Inference Using Time-Domain Neurons

You Only Spike Once: Improving Energy-Efficient Neuromorphic Inference to ANN-Level Accuracy

Improving classification accuracy of feedforward neural networks for spiking neuromorphic chips

Optimizing Power Efficiency: Converting Recurrent Neural Networks to Spiking Neural Networks for Time-Domain Analysis

Spike Trains Encoding and Threshold Rescaling Method for Deep Spiking Neural Networks

PT-Spike: A Precise-Time-Dependent Single Spike Neuromorphic Architecture with Efficient Supervised Learning

30.2 A 22nm 0.26nW/Synapse Spike-Driven Spiking Neural Network Processing Unit Using Time-Step-First Dataflow and Sparsity-Adaptive In-Memory Computing

CMOS-based area-and-power-efficient neuron and synapse circuits for time-domain analog spiking neural networks

Robust and accelerated single-spike spiking neural network training with applicability to challenging temporal tasks

A Time-to-first-spike Coding and Conversion Aware Training for Energy-Efficient Deep Spiking Neural Network Processor Design

Improving energy efficiency and classification accuracy of neuromorphic chips by learning binary synaptic crossbars

Recent Advances in Scalable Energy-Efficient and Trustworthy Spiking Neural networks: from Algorithms to Technology

Spiking neural networks for nonlinear regression

Advancing Spatio-Temporal Processing in Spiking Neural Networks through Adaptation

Training a General Spiking Neural Network with Improved Efficiency and Minimum Latency

High-performance deep spiking neural networks with 0.3 spikes per neuron

Enabling Deep Spiking Neural Networks with Hybrid Conversion and Spike Timing Dependent Backpropagation

Learning Improvement of Spiking Neural Networks with Dynamic Adaptive Hyperparameter Neurons

Conversion of Artificial Recurrent Neural Networks to Spiking Neural Networks for Low-power Neuromorphic Hardware