Repository files navigation

This is an implementation of Olshausen and Field's sparse coding paper in PyTorch. Iterative Shrinkage/Thresholding Algorithm(ISTA) is used to fit neuronal responses for the input. Gradients for receptive fields are calculated through PyTorh's autogradfeature.

To run the program:

cdsrc/scriptspythontrain.py

To see a list of available hyperparameters to change:

pythontrain.py-h

A checkpoint of the model is saved every 10 epochs totrained_models. To see the tensorboard logs:

tensorboard--logdir=runs

• Fast-ISTA

• Olshausen, B. A., & Field, D. J. (1996). Emergence of simple-cell receptive field properties by learning a sparse code for natural images. Nature, 381(6583), 607–609.https://doi.org/10.1038/381607a0
• IMAGES.mat is downloaded from Olshausen's original Matlab implementation website:http://www.rctn.org/bruno/sparsenet/