Framework to train and simulate a spiking neural network (SNN) using Brian2, that implements a learning rule that combines local learning rules with a global feedback mechanism to support strong, contrasty (activity) pattern formation. Such global, "unguided" pattern induction (e.g., through neuromodulatory signals) might lead to easily discriminable output activations even if the network is trained via local (biological-plausible) mechanisms.
The simulation is written in Python 3.
Modules, so far, that have to be installed in order to run the simulation:
- brian2
- numpy
- sklearn
- matplotlib
Additionally, one has to download the datasets for which one ones to run the simulation. Please refer to the dedicated README files in the data folder for the desired dataset for instructions on how to get the dataset.
If all prerequisites are met, one has to configure the simulation in the file src/configuration.py. The file contains extensive comments to guide one through the configuration process.
Finally, one can run the simulation by hitting
python3 main.py
in the src directory.
Note, if cython and/or no c++ compiler is installed, one has to change the target preference in the file src/brian_preferences from cython to numpy.