IM-Loss: Information Maximization Loss for Spiking Neural Networks

Authors: Yufei Guo, Yuanpei Chen, Liwen Zhang, Xiaode Liu, Yinglei Wang, Xuhui Huang, Zhe Ma

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on both popular non-spiking static and neuromorphic datasets show that the SNN models trained by our method outperform the current state-of-the-art algorithms.
Researcher Affiliation Collaboration Yufei Guo , Yuanpei Chen , Liwen Zhang, Xiaode Liu, Yinglei Wang, Xuhui Huang, Zhe Ma Intelligent Science & Technology Academy of CASIC yfguo@pku.edu.cn, rop477@163.com, mazhe_thu@163.com
Pseudocode Yes The algorithm of the training process of our method is presented in Appendix A.2.
Open Source Code No The paper does not contain an explicit statement about the release of source code for the methodology described, nor does it provide a link to a code repository.
Open Datasets Yes The experiments include widely-used network structures including spiking CIFARNet [33], Res Net-19 [38], modified VGG-16 [26], and Res Net-34 [12] for both popular non-spiking static and neuromorphic datasets: CIFAR10/100, Image Net (ILSVRC12) and CIFAR10-DVS [18].
Dataset Splits Yes The experiments include widely-used network structures including spiking CIFARNet [33], Res Net-19 [38], modified VGG-16 [26], and Res Net-34 [12] for both popular non-spiking static and neuromorphic datasets: CIFAR10/100, Image Net (ILSVRC12) and CIFAR10-DVS [18].
Hardware Specification No The paper does not provide specific details regarding the hardware used for running experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions general machine learning frameworks like Pytorch and Tensorflow but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes We set λ as 2 in this paper... In practice, we set Kmin = 100 and Kmax = 101... Both networks were trained with a timestep of 4 and without normalization... For being friendly with neuromorphic hardware, the max-pooling layer was replaced with the average-pooling layer in the used network architectures.