Bridging the Gap between ANNs and SNNs by Calibrating Offset Spikes

Authors: Zecheng Hao, Jianhao Ding, Tong Bu, Tiejun Huang, Zhaofei Yu

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experimental results show that our proposed method achieves state-of-the-art performance on CIFAR10, CIFAR-100, and Image Net datasets. To the best of our knowledge, this is the first time an ANN-SNN conversion has been shown to simultaneously achieve high accuracy and ultralow latency on complex datasets. ... We evaluate our methods on CIFAR-10/100 and Image Net datasets.
Researcher Affiliation Academia Zecheng Hao1, Jianhao Ding1, Tong Bu1, Tiejun Huang1,2 & Zhaofei Yu1,2 1 School of Computer Science, Peking University 2 Institute for Artificial Intelligence, Peking University
Pseudocode Yes A.6 PSEUDO-CODE FOR OVERALL ALGORITHM FLOW: Algorithm 1 Algorithm for ANN-SNN conversion.
Open Source Code Yes Code is available at https://github.com/hzc1208/ANN2SNN_COS.
Open Datasets Yes We choose image classification datasets to validate the effectiveness and performance of our proposed methods, including CIFAR-10 (Le Cun et al., 1998), CIFAR-100 (Krizhevsky et al., 2009) and Image Net (Deng et al., 2009) datasets.
Dataset Splits Yes We choose Stochastic Gradient Descent optimizer (Bottou, 2012) and Cosine Annealing scheduler (Loshchilov & Hutter, 2017) to train ANN models for 300 epochs. For CIFAR-10/100, the value of weight decay is set to 5e-4, and the initial learning rates are 0.1 and 0.02, respectively. For Image Net, we set the initial learning rate as 0.1 and weight decay as 1e-4. In addition, we adopt data-augmentation techniques (De Vries & Taylor, 2017; Cubuk et al., 2019; Li et al., 2021) to further improve the performance of the models.
Hardware Specification No The paper does not provide specific details about the hardware used for running its experiments, such as GPU models, CPU types, or memory specifications.
Software Dependencies No The paper mentions software components like 'Stochastic Gradient Descent optimizer' and 'Cosine Annealing scheduler' but does not specify their version numbers. No other software or library names with specific versions are provided.
Experiment Setup Yes For CIFAR-10/100, the value of weight decay is set to 5e-4, and the initial learning rates are 0.1 and 0.02, respectively. For Image Net, we set the initial learning rate as 0.1 and weight decay as 1e-4. ... We choose Stochastic Gradient Descent optimizer (Bottou, 2012) and Cosine Annealing scheduler (Loshchilov & Hutter, 2017) to train ANN models for 300 epochs. For the setting of the hyperparameter ρ, we set ρ = 4 for CIFAR-10/100 and ρ = 8 for Image Net if there are no special instructions.