Search papers, labs, and topics across Lattice.
This paper introduces the Sparse-Activation-ReLU (SAR) layer, a novel approach that enhances activation sparsity without relying on surrogate-gradient training, thus improving efficiency for edge-deployable virtual sensing. By integrating SAR within a trunk-based NOMAD architecture, the authors achieve over a fivefold enhancement in the Latency-Error-Energy (LEE) metric compared to traditional Variable Spiking Neuron (VSN) and Leaky Integrate-and-Fire (LIF) models. Additionally, the implementation of synthetic knowledge distillation further reduces the LEE score by more than twofold, demonstrating significant advancements in energy-efficient real-time sensing applications.
Achieving over five times improvement in Latency-Error-Energy metrics could redefine efficiency standards for edge-deployable virtual sensing systems.
Virtual sensing enables digital twins and safety-critical systems to reconstruct and forecast spatial-temporal physics in real time. However, conventional computational and data-driven methods often face challenges in generalization, latency, and energy efficiency for edge deployment. Neural operators offer a promising alternative but remain reliant on power-intensive hardware. Spiking neurons and neuromorphic computing can improve efficiency, yet surrogate-gradient training and multi-step spiking introduce convergence and latency challenges. We propose the Sparse-Activation-ReLU (SAR) layer, a single-step alternative that promotes activation sparsity without surrogate-gradient training while remaining compatible with event-based computing. Within a trunk-based NOMAD architecture, SAR achieves over a fivefold improvement in the combined Latency-Error-Energy (LEE) metric compared with Variable Spiking Neuron (VSN) and Leaky Integrate-and-Fire (LIF) implementations. We further analyze spiking entropy and feature usage and introduce synthetic knowledge distillation, reducing the LEE score by more than twofold. Finally, we improve VSN through a ReLU-based spiking loss and graph-neighbor thresholding. On the Heat Exchanger dataset, these approaches reduce L2 error by more than twofold and nearly sevenfold, respectively, while reducing spiking and spatial aggregation. Overall, the work presented is a step towards energy-efficient virtual sensing by providing an alternative framework that can be positioned towards neuromorphic or other edge device integration that can be a gold standard to compare latency, energy, and error performance for future efficient designs that are sparsity or brain-inspired spiking based.
