E-I-SNN

Codebase for the work Biologically-Informed Excitatory and Inhibitory Balance for Robust Spiking Neural Network Training

Simulation workflow

Due to the massive workload of processing hundreds of training sessions (and then multiple layers of processing) this workflow is designed to enable massive batch computing on a SLURM system (which was used here at GWU).

The overview is the following:

1. Generate the shell scripts with each script corresponding to a single SLURM job which may have multiple training sessions (shellScripts/generateShellScripts.py)
2. Submit these scripts with a correspondingly generated runAll.sh script which just submits all of these jobs with a single run
3. Once these trainings have finished, follow similar steps for generating reports (shellScripts/generateShellScripts_Reports.py) and submitting the jobs to generate this.
4. We can aggregate the results into combined csv files using (scripts/generateCSV_SR.py)
5. These final csvs are what the final jupyter notebooks use for generating the corresponding figures

For additional information

Running of each portion of the process is parallelized by making all configuration settings accessible through command line args. For additional information check the --help flag with the fundamental scripts of this repository to see what parameters are accessible.

Key scripts to check:

• scripts/excInhTraining.py
• scripts/generateReport.py

Available data

Included here are the final csv files used in the larger/key figures of the paper. Additional data (in a raw format) is available upon request due to the large nature of the data underneath (which would drastically bloat the repository)

Additional Support

Please file an issue as needed for additional support, questions, or suggestions for code reorganization