Spike frequency adaptation supports network computations on temporally dispersed information

This repository provides a tensorflow library and a tutorial to train a recurrent spiking neural network of adaptive neurons (ALIF, neurons with spike frequency adaptation). The adaptation is implemented as adaptive threshold. The scripts contained reproduce many results from the paper [1].

[1] Spike frequency adaptation supports network computations on temporally dispersed information
Darjan Salaj, Anand Subramoney, Ceca Kraišniković, Guillaume Bellec, Robert Legenstein, Wolfgang Maass

Reproducing results from [1]

Python scripts are in the bin directory, and the notebooks are in the root.

• Figure 1B: constant current response of ALIF neuron 560_fig1_ALIF_step_current_response.py
• Figure 1E: 2 neuron STORE-RECALL tutorial_storerecall_2neuron_solution.py and plot_storerecall.ipynb
• Figure 2A: 20-dim STORE-RECALL tutorial_extended_storerecall_with_LSNN.py and plot_ext_storerecall.ipynb
• Figure 2B: STORE-RECALL comparison of mechanisms tutorial_storerecall_with_LSNN.py and tutorial_storerecall_with_STP.py
• Figure 2C: sMNIST comparison of mechanisms tutorial_sequential_mnist_with_LSNN.py and tutorial_sequential_mnist_with_STP.py
• Table 1: tutorial_storerecall_with_LSNN.py
• Suppl. Figure S1: intrinsic time scale measurements autocorr.py
• Suppl. Figure S2: constant current response of other slow mechanisms other_mechanisms_dynamics_plot.ipynb and nALIF_current_response.py
• Suppl. Figure S4: Delayed-memory XOR tutorial_temporalXOR_with_LSNN.py
• Suppl. Figure S8: Adaptation index plot_adaptation_index_dist.ipynb

Running the scripts

Enter the root directory of the repository and run the training scripts as following:

PYTHONPATH=. python3 bin/tutorial_storerecall_with_LSNN.py --reproduce=560_ALIF

Troubleshooting

If the scripts fail with the following error: Illegal instruction (core dumped)

It is most probably due to the lack of AVX instructions on the machine you are using. A known workaround is to reinstall the LSNN package with older tensorflow version (1.5). Change requirements.txt to contain:

tensorflow==1.5