ECC-SNN: Cost-Effective Edge-Cloud Collaboration for Spiking Neural Networks

TL;DR

ECC-SNN addresses the high energy and latency costs of traditional edge-cloud setups by integrating energy-efficient edge SNNs with cloud-based ANNs. It achieves this through a threefold strategy: (i) a joint ANN-SNN training regime that distills knowledge from a cloud model into an edge SNN, (ii) an ambiguity-aware collaborative inference that offloads only uncertain inputs to the cloud, and (iii) on-device incremental learning that continuously adapts the edge model without excessive cloud communication. The framework demonstrates an average accuracy gain of 4.15% and substantial reductions in energy (79.4%) and latency (39.1%) across four datasets, validating its effectiveness in dynamic IoT environments. By balancing edge processing with selective cloud assistance, ECC-SNN reduces cloud dependence while maintaining high accuracy, enabling cost-effective, responsive on-device intelligence for resource-constrained systems.

Abstract

Most edge-cloud collaboration frameworks rely on the substantial computational and storage capabilities of cloud-based artificial neural networks (ANNs). However, this reliance results in significant communication overhead between edge devices and the cloud and high computational energy consumption, especially when applied to resource-constrained edge devices. To address these challenges, we propose ECC-SNN, a novel edge-cloud collaboration framework incorporating energy-efficient spiking neural networks (SNNs) to offload more computational workload from the cloud to the edge, thereby improving cost-effectiveness and reducing reliance on the cloud. ECC-SNN employs a joint training approach that integrates ANN and SNN models, enabling edge devices to leverage knowledge from cloud models for enhanced performance while reducing energy consumption and processing latency. Furthermore, ECC-SNN features an on-device incremental learning algorithm that enables edge models to continuously adapt to dynamic environments, reducing the communication overhead and resource consumption associated with frequent cloud update requests. Extensive experimental results on four datasets demonstrate that ECC-SNN improves accuracy by 4.15%, reduces average energy consumption by 79.4%, and lowers average processing latency by 39.1%.

Figures (8)

• Figure 1: Overview of the proposed ECC-SNN. In all three stages, the cloud-based ANN model directly or indirectly supports the edge SNN model in preparation, inference, and adaptive updates.
• Figure 2: Optional feature distillation in the joint training approach. $T$ is the number of time steps of the features out of the last SNN layer overlapping with the corresponding ANN layer.
• Figure 3: Case Study: a spiking VGG-9 model learned with CIFAR-10 limited to samples labeled 1 and 6. The entropy distributions and corresponding cloud upload rates for each label are derived from each test sample's model output predictive distribution.
• Figure 4: Changing patterns of accuracy performance and CUR as more classes learned, with a fixed filtering threshold $\delta$=0.3.
• Figure 5: Average accuracy of edge SNN w.r.t. different IL methods.