TC-LIF

Two-Compartment Leaky Integrate-and-Fire Model

The source code for paper: TC-LIF: A Two-Compartment Spiking Neuron Model for Long-term Sequential Modelling, AAAI 2024

To view our paper, please refer: TC-LIF

Method

Dependency

The denpendencies and versions are listed below:

python                  3.8.13
CUDA                    11.7.1
torch                   1.13.0
torchaudio              0.13.0
torchvision             0.14.0
numpy                   1.23.3
urllib3                 1.26.12
spikingjelly            0.0.0.0.13
librosa                 0.7.1
Werkzeug                2.0.3
h5py                    3.10.0

Reproduce

S-MNIST

# recurrent: 64-256-256-10 paras: 155.1K
python MNIST/mn_train.py --neuron tclif --beta1 1.5 --beta2 -0.5 --task SMNIST --threshold 1.0 --gamma 0.5 --sg triangle --network fb --ind 1
# feedforward: 64-256-256-10 paras: 63.6K
python MNIST/mn_train.py --neuron tclif --beta1 0. --beta2 0. --task SMNIST --threshold 1.0 --gamma 0.5 --sg triangle --network ff --ind 1

PS-MNIST

# recurrent: 64-256-256-10 paras: 155.1K
python MNIST/mn_train.py --neuron tclif --beta1 -1.4 --beta2 1.4 --task PSMNIST --threshold 1.8 --gamma 0.5 --sg triangle --network fb --ind 1
# feedforward: 64-256-256-10 paras: 63.6K
python MNIST/mn_train.py --neuron tclif --beta1 0. --beta2 0. --task PSMNIST --threshold 1.5 --gamma 0.7 --sg triangle --network ff --ind 1

GSC

# recurrent: 40-300-300-12 paras: 196.5K
python GSC-v2/gg12_train.py --neuron tclif --beta_init1 1.4 --beta_init2 1.4 --threshold 1.25 --gamma 0.7 --sg triangle --network fb --version v2 --drop 0.3
# feedforward: 40-300-300-12 paras: 106.2K
python GSC-v2/gg12_train.py --neuron tclif --beta_init1 1.4 --beta_init2 1.4 --threshold 1.2 --gamma 0.6 --sg triangle --network ff --version v2 --drop 0.3

SHD

The SHD and SSC can be downloaded in zenkelab.

# recurrent: 700-128-128-20 paras: 141.8K
python SHD-SSC/main.py --neuron tclif --task SHD --beta1 0. --beta2 0. --threshold 1.5 --gamma 0.5 --sg triangle --network fb 
# feedforward: 700-128-128-20 paras: 108.8K
python SHD-SSC/main.py --neuron tclif --task SHD --beta1 0. --beta2 0. --threshold 1.5 --gamma 0.5 --sg triangle --network ff

SSC

# recurrent: 700-128-128-135 paras: 110.8K
python SHD-SSC/main.py --neuron tclif --task SSC --beta1 0. --beta2 0. --threshold 1.5 --gamma 0.5 --sg triangle --network fb
# feedforward: 700-128-135 paras: 110.8K
python SHD-SSC/main.py --neuron tclif --task SSC --beta1 0. --beta2 0. --threshold 1.5 --gamma 0.5 --sg triangle --network ff

Node1

# S-MNIST feedforward
python mn_train.py --neuron node1 --task SMNIST --threshold 1.0 --gamma 0.5 --sg triangle --network ff --ind 1

Node2

# S-MNIST feedforward
CUDA_VISIBLE_DEVICES=1 python mn_train.py --neuron node2 --threshold1 0.8 --threshold2 1.2 --task SMNIST --sg triangle --network ff --ind 1

Node3

# S-MNIST feedforward
CUDA_VISIBLE_DEVICES=2 python mn_train.py --neuron node3 --gamma1 0.5 --gamma2 0.7 --task SMNIST --sg triangle --network ff --ind 1

Node4

# S-MNIST feedforward
CUDA_VISIBLE_DEVICES=3 python mn_train.py --neuron node4 --threshold1 0.8 --threshold2 1.2 --task SMNIST --sg triangle --network ff --ind 1