SPARTA: Advancing Sparse Attention in Spiking Neural Networks via Spike-Timing-Based Prioritization

2025-08-03

AI Summary

Existing spiking neural networks (SNNs) predominantly rely on rate coding, neglecting the rich temporal dynamics encoded in precise spike timing—such as spike latency, inter-spike intervals, and firing patterns—leading to computational redundancy and limited representational capacity. To address this, we propose SPARTA, the first framework to explicitly incorporate fine-grained spike timing into attention mechanisms. SPARTA introduces time-aware priority gating and resource-adaptive sparse selection, jointly optimized with heterogeneous neuron dynamics modeling. Evaluated on DVS-Gesture, CIFAR10-DVS, and static CIFAR-10 (with event-to-frame conversion), SPARTA achieves 98.78%, 83.06%, and 95.3% accuracy, respectively, while attaining 65.4% synaptic sparsity. These results significantly outperform state-of-the-art SNNs, establishing a novel paradigm for temporally driven, brain-inspired attention that balances efficiency and expressivity.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Machine Learning: Hardware-aware ML

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Web Mining and Content Analysis: Models for Web evolution

Abstract

Current Spiking Neural Networks (SNNs) underutilize the temporal dynamics inherent in spike-based processing, relying primarily on rate coding while overlooking precise timing information that provides rich computational cues. We propose SPARTA (Spiking Priority Attention with Resource-Adaptive Temporal Allocation), a framework that leverages heterogeneous neuron dynamics and spike-timing information to enable efficient sparse attention. SPARTA prioritizes tokens based on temporal cues, including firing patterns, spike timing, and inter-spike intervals, achieving 65.4% sparsity through competitive gating. By selecting only the most salient tokens, SPARTA reduces attention complexity from O(N^2) to O(K^2) with k << n, while maintaining high accuracy. Our method achieves state-of-the-art performance on DVS-Gesture (98.78%) and competitive results on CIFAR10-DVS (83.06%) and CIFAR-10 (95.3%), demonstrating that exploiting spike timing dynamics improves both computational efficiency and accuracy.

Problem

• Enhances SNNs by utilizing precise spike timing dynamics
• Reduces attention complexity from O(N^2) to O(K^2)
• Improves computational efficiency and accuracy in neural networks

Innovation

• Leverages spike-timing for efficient sparse attention
• Reduces complexity via competitive gating and prioritization
• Achieves high accuracy with dynamic token selection