Spiking neural network simulator which allows to simulate neuronal cultures placed over Multi-Electrode Arrays (MEAs).
MEAs can be used to stimulate and/or record the activity of the neuronal culture. Thus, these devices might enable the use of neuronal cultures as biological computing systems. For example, it is possible to stimulate the neurons with input patterns such as digits, record the resulting response and use it to classify these patterns. Biological computation is attractive because it has the potential to solve complex tasks with a very limited energy consumption (as it happens, for example, in the brain).
Our goal is to explore whether it is possible to train neural cultures on MEA devices to solve such kind of tasks. This simulator can be used for finding an appropriate set of network parameters in order to solve a given task. Once the appropriate parameters are known, it is then possible, through modern neurobiology techniques, to produce equivalent biological cultures.
A fundamental aspect of this work is to show the potential of the interplay between Computer Science and Neurobiology.
Link to the paper: https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9441166&casa_token=gkxO3Q1EfBoAAAAA:EZZ_j6GqcVivb7NH-a6N4jDAUeiYFGsg9ub2Ju1mNeL4_00P3RI--YnIQXoL4zHY8U8FWJTokt8&tag=1
The simulator can be used as follows:
In the file configs.py it is possible to specify custom experimental
configurations, defining the various parameters of the simulation.
The script runexp.py can be used to launch an experiment session.
Experiments are performed on the MNIST dataset for digit recognition.
The script bioexp.py is another experiment based on the following
work on neuronal cultures: https://ieeexplore.ieee.org/document/1396377
We simulated this experiment in order to obtain an estimate for the
parameters to be used as a starting point for successive simulations.
Once you have set up your configuration, you can launch a training
session with the command:
python runexp.py --config <config name> --mode train
Where <config name> is the name of one of the training configurations
in the config.py file.
To test the resulting model, type:
python runexp.py --config <config name> --mode test
REQUIREMENTS:
- python 3.6
- matplotlib 3.1.1
- numpy 1.17.2
- scipy 1.4.1
- pytorch 1.4.0
- torchvision 0.5.0
- bindsnet 0.2.6