This repository presents an implementation of the Neural Module Network framework leveraging parallel computing and GPU-accelerated computation. The demonstration employs the Facebook social network dataset, a high-dimensional corpus comprising 20,162 samples characterized by 4,714 distinct features per instance.

To begin, please run the script Neural_Modules.m.

For tuning parameters, refer to the file located at ./NN/nnsetup.m.

Note that:
1. nn.top_k corresponds to the top k weight for gamma mentioned in the paper.
2. nn.alpha corresponds to the alpha parameter mentioned in the paper.

