Discretized Quadratic Integrate-and-Fire Neuron Model for Direct Training of Spiking Neural Networks

ICLR 2025 Conference Submission1187 Authors

Keywords: Machine Learning, Neuromorphic Computing, Deep Learning, Computer Vision

Abstract: Spiking Neural Networks (SNNs) are a promising alternative to traditional artificial neural networks, offering significant energy-saving potential. Conventional SNN approaches typically utilize the Leaky Integrate-and-Fire (LIF) neuron model, where voltage decays linearly, decreasing proportionally to its current value. However, this linear decay can inadvertently increase energy consumption and reduce model performance due to extraneous spiking activity. To address these limitations, we introduce the discretized Quadratic Integrate-and-Fire (QIF) neuron model, which applies a non-linear transformation to the voltage proportional to its magnitude. The QIF neuron model achieves substantial energy reductions, ranging from $1.43 - 4.21\times$ compared to the LIF neuron model. On static datasets (CIFAR-10, CIFAR-100) and neuromorphic datasets (CIFAR-10 DVS, N-Caltech-101, N-Cars, DVS128-Gesture), the QIF neuron model demonstrates competitive performance and improved accuracy over state-of-the-art results. Furthermore, the QIF neuron model produces smoother loss landscapes and larger local minima, leading to faster training convergence. Our findings suggest that the QIF neuron model offers a promising alternative to the widely adopted LIF neuron model.

Primary Area: other topics in machine learning (i.e., none of the above)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 1187