# Collapse or Preserve: Data-Dependent Temporal Aggregation for Spiking Neural Network Acceleration

arXiv: https://arxiv.org/abs/2603.13810 · HTML: https://mlsanity.com/paper/2603.13810/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 13 of its 24 references that are in the map.
- First posted on arXiv 14 Mar 2026.

## Builds on (most cited first, 13 of 13)

- [Surrogate Gradient Learning in Spiking Neural Networks](https://mlsanity.com/paper/1901.09948/) (2019)
- [Spatio-Temporal Backpropagation for Training High-performance Spiking Neural Networks](https://mlsanity.com/paper/1706.02609/) (2017)
- [Incorporating Learnable Membrane Time Constant to Enhance Learning of Spiking Neural Networks](https://mlsanity.com/paper/2007.05785/) (2020)
- [Going Deeper With Directly-Trained Larger Spiking Neural Networks](https://mlsanity.com/paper/2011.05280/) (2020)
- [Training Spiking Neural Networks Using Lessons From Deep Learning](https://mlsanity.com/paper/2109.12894/) (2021)
- [Temporal Efficient Training of Spiking Neural Network via Gradient Re-weighting](https://mlsanity.com/paper/2202.11946/) (2022)
- [SpikingJelly: An open-source machine learning infrastructure platform for spike-based intelligence](https://mlsanity.com/paper/2310.16620/) (2023)
- [The Hardware Lottery](https://mlsanity.com/paper/2009.06489/) (2020)
- [Temporal-wise Attention Spiking Neural Networks for Event Streams Classification](https://mlsanity.com/paper/2107.11711/) (2021)
- [Revisiting Batch Normalization for Training Low-latency Deep Spiking Neural Networks from Scratch](https://mlsanity.com/paper/2010.01729/) (2020)
- [Pruning of Deep Spiking Neural Networks through Gradient Rewiring](https://mlsanity.com/paper/2105.04916/) (2021)
- [TCJA-SNN: Temporal-Channel Joint Attention for Spiking Neural Networks](https://mlsanity.com/paper/2206.10177/) (2022)
- [Mamba-Spike: Enhancing the Mamba Architecture with a Spiking Front-End for Efficient Temporal Data Processing](https://mlsanity.com/paper/2408.11823/) (2024)

## Abstract (arXiv)

Spike sparsity is widely believed to enable efficient spiking neural network (SNN) inference on GPU hardware. We demonstrate this is an illusion: five distinct sparse computation strategies on Apple M3 Max all fail to outperform dense convolution, because SIMD architectures cannot exploit the fine-grained, unstructured sparsity of i.i.d. binary spikes. Instead, we propose Temporal Aggregated Convolution (TAC), which exploits convolution linearity to pre-aggregate K spike frames before a single convolution call, reducing T calls to T/K. On rate-coded data, TAC achieves 13.8times speedup with +1.6% accuracy on MNIST and +5.4% on Fashion-MNIST -- a simultaneous improvement in both speed and accuracy. However, on event-based data where the temporal dimension carries genuine motion information, TAC's temporal collapse is harmful. We therefore introduce TAC-TP (Temporal Preservation), which shares each group's convolution output across K independent LIF steps, preserving full temporal resolution for downstream layers. On DVS128-Gesture, TAC-TP achieves 95.1% accuracy (vs. 96.3% baseline) with 50% fewer convolution calls, while standard TAC drops to 91.3%. Our key finding is that the optimal temporal aggregation strategy is data-dependent: collapse the temporal dimension for rate-coded data (noise reduction) but preserve it for event data (information retention). Speedup is hardware-agnostic: TAC achieves 11.0times on NVIDIA V100, confirming the mechanism transfers across GPU architectures. All operators in the mlx-snn library are open source.