LightSNN: Lightweight Architecture Search for Sparse and Accurate Spiking Neural Networks

Spiking Neural Networks (SNNs) are highly regarded for their energy efficiency, inherent activation sparsity, and suitability for real-time processing in edge devices. However, most current SNN methods adopt architectures resembling traditional artificial neural networks (ANNs), leading to suboptimal performance when applied to SNNs. While SNNs excel in energy efficiency, they have been associated with lower accuracy levels than traditional ANNs when utilizing conventional architectures. In response, in this work we present LightSNN, a rapid and efficient Neural Network Architecture Search (NAS) technique specifically tailored for SNNs that autonomously leverages the most suitable architecture, striking a good balance between accuracy and efficiency by enforcing sparsity. Based on the spiking NAS network (SNASNet) framework, a cell-based search space including backward connections is utilized to build our training-free pruning-based NAS mechanism. Our technique assesses diverse spike activation patterns across different data samples using a sparsity-aware Hamming distance fitness evaluation. Thorough experiments are conducted on both static (CIFAR10 and CIFAR100) and neuromorphic datasets (DVS128-Gesture). Our LightSNN model achieves state-of-the-art results on CIFAR10 and CIFAR100, improves performance on DVS128Gesture by 4.49%, and significantly reduces search time, most notably offering a 98x speedup over SNASNet and running 30% faster than the best existing method on DVS128Gesture.

Open-Set Gait Recognition from Sparse mmWave Radar Point Clouds

The adoption of Millimeter-Wave (mmWave) radar devices for human sensing, particularly gait recognition, has recently gathered significant attention due to their efficiency, resilience to environmental conditions, and privacy-preserving nature. In this work, we tackle the challenging problem of Open-set Gait Recognition (OSGR) from sparse mmWave radar point clouds. Unlike most existing research, which assumes a closed-set scenario, our work considers the more realistic open-set case, where unknown subjects might be present at inference time, and should be correctly recognized by the system. Point clouds are well-suited for edge computing applications with resource constraints, but are more significantly affected by noise and random fluctuations than other representations, like the more common micro-Doppler signature. This is the first work addressing open-set gait recognition with sparse point cloud data. To do so, we propose a novel neural network architecture that combines supervised classification with unsupervised reconstruction of the point clouds, creating a robust, rich, and highly regularized latent space of gait features. To detect unknown subjects at inference time, we introduce a probabilistic novelty detection algorithm that leverages the structured latent space and offers a tunable trade-off between inference speed and prediction accuracy. Along with this paper, we release mmGait10, an original human gait dataset featuring over five hours of measurements from ten subjects, under varied walking modalities. Extensive experimental results show that our solution attains F1-Score improvements by 24% over state-of-the-art methods, on average, and across multiple openness levels.

Learned Spike Encoding of the Channel Response for Low-Power Environment Sensing

Radio Frequency (RF) sensing holds the potential for enabling pervasive monitoring applications. However, modern sensing algorithms imply complex operations, which clash with the energy-constrained nature of edge sensing devices. This calls for the development of new processing and learning techniques that can strike a suitable balance between performance and energy efficiency. Spiking Neural Networks (SNNs) have recently emerged as an energy-efficient alternative to conventional neural networks for edge computing applications. They process information in the form of sparse binary spike trains, thus potentially reducing energy consumption by several orders of magnitude. Their fruitful use for RF signal processing critically depends on the representation of RF signals in the form of spike signals. We underline that existing spike encoding algorithms to do so generally produce inaccurate signal representations and dense (i.e., inefficient) spike trains. In this work, we propose a lightweight neural architecture that learns a tailored spike encoding representations of RF channel responses by jointly reconstructing the input and its spectral content. By leveraging a tunable regularization term, our approach enables fine-grained control over the performance-energy trade-off of the system. Our numerical results show that the proposed method outperforms existing encoding algorithms in terms of reconstruction error and sparsity of the obtained spike encodings.

Enhanced Attention-Based Unrolling for Sparse Sequential micro-Doppler Reconstruction

The reconstruction of micro-Doppler signatures of human movements is a key enabler for fine-grained activity recognition with radio-frequency sensing. In this work, we focus on Joint Communication and Sensing (JCS) systems where, unlike in dedicated radar sensing systems, a suitable tradeoff between sensing accuracy and communication overhead has to be attained. It follows that the micro-Doppler has to be reconstructed from sparse and noisy channel estimates obtained from communication packets, limiting as much as possible the transmission of additional probing signals for the purpose of sensing. Existing approaches exploit compressed sensing, but produce very poor reconstructions when only a few channel measurements are available, which is often the case in real communication patterns. In addition, the large number of iterations they need to converge hinders their use in real-time systems. Here, we present STAR, a lightweight neural network that combines a single unrolled iterative hard-thresholding layer with an attention mechanism. Our new approach exploits the temporal correlation of the micro-Doppler to accurately reconstruct microDoppler sequences from human movement even from very sparse channel measurements. In doing so, it combines model-based and data-driven approaches into an interpretable and low-complexity architecture, which is amenable to real-time implementations. We evaluate STAR on a public JCS dataset of 60 GHz IEEE 802.11ay channel measurements of human activity traces. Experimental results show that it substantially outperforms state-of-the-art solutions in terms of the reconstructed microDoppler quality. Remarkably, STAR enables human activity recognition with satisfactory accuracy even with 90%-sparse channel measurements, for which existing techniques fail.