Adaptive Sparse Connectivity for Neural Networks in PyTorch
A PyTorch implementation of Scalable Training of Artificial Neural Networks with Adaptive Sparse Connectivity inspired by Network Science by Mocanu et al. (https://arxiv.org/abs/1707.04780) Uses sparse data structures. Not super fast yet, but less memory-intensive than the masked dense weight matrices used in the proof-of-concept code released with the paper.
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
synapses-0.0.1.tar.gz (3.4 kB view hashes)