Exploring Spiking Neural Networks for Binary Classification in Multivariate Time Series at the Edge

TL;DR

This work addresses binary classification in multivariate time series at the edge by developing a general, hardware-aware framework that evolves compact, stateful SNNs via the Evolutionary Optimization of Neuromorphic Systems (EONS) algorithm. Spikes are encoded from inputs and predictions arise from thresholding an output neuron's spike count, with MCC used as the fitness during evolution and simple voting ensembles to boost robustness. The framework, validated on low-SNR gamma-ray source detection and EEG seizure detection, yields SNNs with as few as 49 neurons that outperform PCA and small DL baselines, while enabling energy-efficient edge deployment on the µCaspian platform (≈2 mW, ~20 ms latency). Ensembling further improves performance (e.g., TPR up to 0.671 at FAR ≤ 1/hr on gamma-ray data; MCC ≈0.71 for EEG trio majority), demonstrating practical impact for real-time, low-power edge AI with generalizability across domains.

Abstract

We present a general framework for training spiking neural networks (SNNs) to perform binary classification on multivariate time series, with a focus on step-wise prediction and high precision at low false alarm rates. The approach uses the Evolutionary Optimization of Neuromorphic Systems (EONS) algorithm to evolve sparse, stateful SNNs by jointly optimizing their architectures and parameters. Inputs are encoded into spike trains, and predictions are made by thresholding a single output neuron's spike counts. We also incorporate simple voting ensemble methods to improve performance and robustness. To evaluate the framework, we apply it with application-specific optimizations to the task of detecting low signal-to-noise ratio radioactive sources in gamma-ray spectral data. The resulting SNNs, with as few as 49 neurons and 66 synapses, achieve a 51.8% true positive rate (TPR) at a false alarm rate of 1/hr, outperforming PCA (42.7%) and deep learning (49.8%) baselines. A three-model any-vote ensemble increases TPR to 67.1% at the same false alarm rate. Hardware deployment on the microCaspian neuromorphic platform demonstrates 2mW power consumption and 20.2ms inference latency. We also demonstrate generalizability by applying the same framework, without domain-specific modification, to seizure detection in EEG recordings. An ensemble achieves 95% TPR with a 16% false positive rate, comparable to recent deep learning approaches with significant reduction in parameter count.

Figures (3)

• Figure 1: Testing dataset ROC curves for the top 5 SNN and reference ARAD and SAD algorithms.
• Figure 2: Spike response of SNN 141 on a time series from the testing dataset, with a synthetic shielded $^{137}$Cs source injected on measured background.
• Figure 3: Recall for the top ensemble and its individual constituents for all non-shielded sources and all sources together (All).