dghernandez / deep-retina

deep-retina is a project to build a convolutional neural network that can predict retinal ganglion cell responses to natural stimuli with high accuracy.

Deep Learning Models of the Retinal Response to Natural Scenes

Deep retina is a project to test to what degree artificial neural networks can predict retinal ganglion cell responses to natural stimuli.

Please see our NIPS paper for more details.

Note that deepretina requires python 3.5 or higher.

Usage

To install the dependencies, run pip install -r requirements.txt. If you run the runme.py script, it will print out a brief overview of the different modules in deepretina (assuming it is able to import everything correctly).

The following is a high level description of the different modules:

• core.py: contains a function for training a deepretina model
• models.py: contains functions for building different kinds of deepretina models (convnets, RNNs, etc.)
• experiments.py: class structure for loading experimental data
• io.py: contains tools for saving model training progress and parameters to disk

A more comprehensive tutorial is in the works.

Contact

Lane McIntosh ([e-mail]) and Niru Maheswaranathan ([e-mail])