Recommend Python version 3.6
pip install -r requirements.txt
To get the results in Supplementary Figures 2.1 and 2.2, you can run the main.m in "time_varied_simulation/fully-connected" or "time_varied_simulation/xor" folder with MATLAB. These two files run the simulation of HH and tv-LIF2HH models for Supplementary Figures 2.1 and 2.2.
matlab main
To get the results in Fig.3d, you can run the main.m in "simplified_simulation" folder with MATLAB. Different types and amplitudes of input signals can be chosen in main.m. This file run the simulation of HH and s-LIF2HH models for Figure 3.
matlab main
For model training, you can run the main.py in "multi-task" folder with Python. Different model names, including "LIF_fc", "HH_fc", "LIF_hh_fc", "4LIF_fc", "ANN", "LIF_conv", "HH_conv", "LIF_hh_conv", "4LIF_conv" and "CNN", can be chosen. This file trains each model and prints the accuracy in the multi-task experiments, for Figure 4c.
cd multi-task
python main.py --model_name LIF_fc
For robustness test, the amplitude of noise can be modified by changing the value of "A". Different model names, including "LIF_fc", "HH_fc", "LIF_hh_fc", "4LIF_fc", "ANN", "LIF_conv", "HH_conv", "LIF_hh_conv", "4LIF_conv" and "CNN", can be chosen. You can run the noise_test.py in "multi-task" folder with Python. This file adds corresponding noises, trains each model and prints the accuracy in the robustness experiments, for Figure 4h.
python noise_test.py --model_name LIF_fc --A 75
For the InvertedPendulum environment, you can run the main.py in "drl_InvertedPendulum" folder with Python. The model name can chosen in "LIF", "HH", "LIF_HH", "4LIF" and "ANN". The results are recorded in the "record/model_name" folder. This file trains each model and records the rewards in the deep reinforcement learning experiment, for Figure 4d (upper one).
cd drl_InvertedPendulum
python main.py --model_name LIF
For the InvertedDoublePendulum environment, you can run the main.py in "drl_InvertedDoublePendulum" folder with Python. The model name can chosen in "LIF", "HH", "LIF_HH", "4LIF" and "ANN". The results are recorded in the "record/model_name" folder. You can add state or reward noise and change the amplitude of noise in the main.py. This file trains each model and records the rewards in the deep reinforcement learning experiment, for Figure 4d (bottom one) and 4i.
cd drl_InvertedDoublePendulum
python main.py --model_name LIF
To measure the mutual information of each network, you can run the main.py in "MI" folder with Python. The model name "LIF" can be changed to "HH" or "LIF_HH". This file calculates the mutual information for Figure 5d.
cd MI
python main.py --model LIF
To measure the FLOPs in each network, you can run the cal_flops.ipynb in "calculate_flops" folder. This file is for Figure 5b.
@article{he2024network,
title={Network model with internal complexity bridges artificial intelligence and neuroscience},
author={He, Linxuan and Xu, Yunhui and He, Weihua and Lin, Yihan and Tian, Yang and Wu, Yujie and Wang, Wenhui and Zhang, Ziyang and Han, Junwei and Tian, Yonghong and others},
journal={Nature Computational Science},
pages={1--16},
year={2024},
publisher={Nature Publishing Group US New York}
}