Overall	Spiking Token Block	Spiking Channel Block	Spiking Patch Encoding

This project provides the official implementation for our paper "Efficient Deep Spiking Multi-Layer Perceptrons with Multiplication-Free Inference". Mixer-SNN is a spiking MLP architecture that uses batch normalization to retain MFI compatibility and introduces a spiking patch encoding layer to reinforce local feature extraction capabilities. Our network secures a top-1 accuracy of 66.39% on the ImageNet-1K dataset, surpassing the directly trained spiking ResNet-34 by 2.67%. An expanded version of our network compares with the performance of the spiking VGG-16 network with a 71.64% top-1 accuracy, all while operating with a model capacity 2.1 times smaller.

git clone https://github.com/BugMaker-Boyan/Mixer-SNN.git
cd Mixer-SNN
conda create -n mixer_snn python=3.9
conda activate mixer_snn
pip install -r requirements.txt

Additionally, install cupy package based on your cuda version.

# For CUDA 11.2 ~ 11.x
pip install cupy-cuda11x

# For CUDA 12.x
pip install cupy-cuda12x

To train a Mixer-SNN-Small model on ImageNet-1K dataset with 8 GPUs, run:

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch \
	--nproc_per_node=8 train.py \
	--model_size small
	--T 4 \
	--cupy \
	--amp \
	--batch-size 64 \
	--model mixer_sparse \
	--epochs 100 \
	--output-dir ./logs \
	--lr 0.1 \
	--lr-scheduler cosa \
	--data imagenet \
	--data-path /data/ILSVRC2012 \
	--opt sgd \
	--lr-warmup-epochs 0 \
	--exp-name Mixer-SNN-Small-ImageNet

To train a Mixer-SNN-Small model on CIFAR10 dataset with 8 GPUs, run:

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch \
	--nproc_per_node=8 train.py \
	--model_size small
	--T 4 \
	--cupy \
	--amp \
	--batch-size 64 \
	--model mixer_sparse \
	--epochs 100 \
	--output-dir ./logs \
	--lr 0.1 \
	--lr-scheduler cosa \
	--data cifar10 \
	--data-path /data/liboyan/ILSVRC2012 \
	--opt sgd \
	--lr-warmup-epochs 0 \
	--exp-name Mixer-SNN-Small-ImageNet

You can watch the training process via tensorboard:

tensorboadr --logdir=./logs

For more details,please refer to our paper Efficient Deep Spiking Multi-Layer Perceptrons with Multiplication-Free Inference.

@article{li2024efficient,
         title={Efficient Deep Spiking Multi-Layer Perceptrons with Multiplication-Free Inference}, 
         author={Boyan Li and Luziwei Leng and Shuaijie Shen and Kaixuan Zhang and Jianguo Zhang and Jianxing Liao and Ran Cheng},
         year={2024},
         journal={IEEE Transactions on Neural Networks and Learning Systems},
         doi={10.1109/TNNLS.2024.3394837}
}