Memory-Efficient aggregations

Memory-Efficient aggregations

Posted by Gabriele Santin on December 17, 2021

Memory-Efficient aggregations

In this tutorial, we discuss how the new version of PyTorch Geometric leverages sparse matrices to improve the computational efficiency and memory footprint of message passing. We review some background on the representation of sparse matrices, have a look at the SparseTensor class from PyTorch Sparse, and analyze the new message_and_aggregate method in PyG.






Download the material of the lecture here.