Linear recurrent neural networks enable powerful long-range sequence modeling with constant memory usage and time-per-token during inference. These architectures hold promise for streaming applications at the edge, but deployment in resource-constrained environments requires hardware-aware optimizations to minimize latency and energy consumption. Unstructured sparsity offers a compelling solution, enabling substantial reductions in compute and memory requirements–when accelerated by compatible hardware platforms. In this paper, we conduct a scaling study to investigate the Pareto front of performance and efficiency across inference compute budgets. We find that highly sparse linear RNNs consistently achieve better efficiency-performance trade-offs than dense baselines, with $2\times$ less compute and $36$% less memory at iso-accuracy. Our models achieve state-of-the-art results on a real-time streaming task for audio denoising. By quantizing our sparse models to fixed-point arithmetic and deploying them on the Intel Loihi 2 neuromorphic chip for real-time processing, we translate model compression into tangible gains of $42\times$ lower latency and $149\times$ lower energy consumption compared to a dense model on an edge GPU. Our findings showcase the transformative potential of unstructured sparsity, paving the way for highly efficient recurrent neural networks in real-world, resource-constrained environments.

Loihi 2 is an asynchronous, brain-inspired research processor that generalizes several fundamental elements of neuromorphic architecture, such as stateful neuron models communicating with event-driven spikes, in order to address limitations of the first generation Loihi. Here we explore and characterize some of these generalizations, such as sigma-delta encapsulation, resonate-and-fire neurons, and integer-valued spikes, as applied to standard video, audio, and signal processing tasks. We find that these new neuromorphic approaches can provide orders of magnitude gains in combined efficiency and latency (energy-delay-product) for feed-forward and convolutional neural networks applied to video, audio denoising, and spectral transforms compared to state-of-the-art solutions.

Spiking Neural Networks (SNNs) are a promising research paradigm for low power edge-based computing. Recent works in SNN backpropagation has enabled training of SNNs for practical tasks. However, since spikes are binary events in time, standard loss formulations are not directly compatible with spike output. As a result, current works are limited to using mean-squared loss of spike count. In this paper, we formulate the output probability interpretation from the spike count measure and introduce spike-based negative log-likelihood measure which are more suited for classification tasks especially in terms of the energy efficiency and inference latency. We compare our loss measures with other existing alternatives and evaluate using classification performances on three neuromorphic benchmark datasets: NMNIST, DVS Gesture and N-TIDIGITS18. In addition, we demonstrate state of the art performances on these datasets, achieving faster inference speed and less energy consumption.

Spiking Neural Networks (SNNs) are bio-inspired networks that process information conveyed as temporal spikes rather than numeric values. An example of a sensor providing such data is the event-camera. It only produces an event when a pixel reports a significant brightness change. Similarly, the spiking neuron of an SNN only produces a spike whenever a significant number of spikes occur within a short period of time. Due to their spike-based computational model, SNNs can process output from event-based, asynchronous sensors without any pre-processing at extremely lower power unlike standard artificial neural networks. This is possible due to specialized neuromorphic hardware that implements the highly-parallelizable concept of SNNs in silicon. Yet, SNNs have not enjoyed the same rise of popularity as artificial neural networks. This not only stems from the fact that their input format is rather unconventional but also due to the challenges in training spiking networks. Despite their temporal nature and recent algorithmic advances, they have been mostly evaluated on classification problems. We propose, for the first time, a temporal regression problem of numerical values given events from an event-camera. We specifically investigate the prediction of the 3- DOF angular velocity of a rotating event-camera with an SNN. The difficulty of this problem arises from the prediction of angular velocities continuously in time directly from irregular, asynchronous event-based input. Directly utilising the output of event-cameras without any pre-processing ensures that we inherit all the benefits that they provide over conventional cameras. That is high-temporal resolution, high-dynamic range and no motion blur. To assess the performance of SNNs on this task, we introduce a synthetic event-camera dataset generated from real-world panoramic images and show that we can successfully train an SNN to perform angular velocity regression.

Configuring deep Spiking Neural Networks (SNNs) is an exciting research avenue for low power spike event based computation. However, the spike generation function is non-differentiable and therefore not directly compatible with the standard error backpropagation algorithm. In this paper, we introduce a new general backpropagation mechanism for learning synaptic weights and axonal delays which overcomes the problem of non-differentiability of the spike function and uses a temporal credit assignment policy for backpropagating error to preceding layers. We describe and release a GPU accelerated software implementation of our method which allows training both fully connected and convolutional neural network (CNN) architectures. Using our software, we compare our method against existing SNN based learning approaches and standard ANN to SNN conversion techniques and show that our method achieves state of the art performance for an SNN on the MNIST, NMNIST, DVS Gesture, and TIDIGITS datasets.

Stability is a key issue during spiking neural network training using SpikeProp. The inherent nonlinearity of Spiking Neuron means that the learning manifold changes abruptly; therefore, we need to carefully choose the learning steps at every instance. Other sources of instability are the external disturbances that come along with training sample as well as the internal disturbances that arise due to modeling imperfection. The unstable learning scenario can be indirectly observed in the form of surges, which are sudden increases in the learning cost and are a common occurrence during SpikeProp training. Research in the past has shown that proper learning step size is crucial to minimize surges during training process. To determine proper learning step in order to avoid steep learning manifolds, we perform weight convergence analysis of SpikeProp learning in the presence of disturbance signals. The weight convergence analysis is further extended to robust stability analysis linked with overall system error. This ensures boundedness of the total learning error with minimal assumption of bounded disturbance signals. These analyses result in the learning rate normalization scheme, which are the key results of this paper. The performance of learning using this scheme has been compared with the prevailing methods for different benchmark data sets and the results show that this method has stable learning reflected by minimal surges during learning, higher success in training instances, and faster learning as well.

Supervised Learning methods for Spiking Neural Network are either able to learn spike train for a single neuron or able to learn first spike in a multilayer feedforward connection setting. The first group of learning methods do not use the computational benefits of hidden layer neuron whereas the second group of learning methods do not exploit the information transfer potential of spike train. Although, there have been few efforts to learn spike train in multilayer feedforward setting for spiking neural networks, the computational cost of these methods increases when spike train is considered for long period. We present spike event based weight update strategy that is able to learn spike train pattern in multilayer feedforward spiking neural network and is efficient and scalable for learning spike train pattern for indefinite period of time. We will compare this method with relevant spiking neural network learning algorithms based on different benchmark datasets and show the efficacy of this event based weight update learning.

Delay learning in SpikeProp is useful because it eliminates the need of redundant synaptic connections in a Spiking Neural Network (SNN). The delay learning enhancement to SpikeProp, however, also inherits the complications present in basic SpikeProp with weight update that obstruct the learning process. To tackle these issues, we perform delay convergence analysis to investigate the conditions required to assure convergence during learning and propose an adaptive learning rate scheme based on it. The comparison of this adaptive learning rate method with the basic fixed learning rate implementation and adaptive weight update only methods via simulations on various standard benchmarks show that it indeed improves the learning convergence and speeds up the learning process.