New Research Advances Energy-Efficient Spiking Neural Networks and Optimization Methods

Researchers develop energy-efficient spiking neural networks for beam prediction and introduce convex optimization techniques for SNN training.

New Research Advances Energy-Efficient Spiking Neural Networks and Optimization Methods

Researchers have published multiple studies advancing spiking neural networks (SNNs), focusing on energy efficiency and optimization techniques.

According to a paper published on arxiv.org, scientists developed SNN architectures for sensing-aided beam prediction in millimeter-wave systems. The research systematically compared SNNs with conventional artificial neural networks (ANNs) using position-only, vision-only, and multimodal inputs. For vision and multimodal sensing, SNNs achieved higher accuracy than ANNs at similar energy costs and required “up to an order of magnitude less energy at a comparable accuracy,” according to the paper. The multimodal SNN achieved 58.08% top-1 accuracy, outperforming the closest-energy ANN by 18.11 percentage points. However, the study found that “SNN accuracy is limited by the binary communication between layers rather than by model size.”

In a separate development, researchers addressed SNN training challenges by formulating SNN optimization as a convex optimization problem, according to arxiv.org. The team showed that “the weight-decay-regularized training problem for a class of threshold RNNs, including leaky integrate-and-fire SNNs, has zero duality gap and the same optimal value as a convex program.” Empirically, the convex variant matched surrogate-gradient training performance on sequential MNIST and outperformed it on addition and first-last-XOR tasks, according to the paper.

Both studies were submitted to arxiv.org in 2026.

spiking neural networksenergy efficiencyoptimizationbeam predictionconvex optimization