Preprint on modeling paths with graph neural networks
Jean-Baptiste Cordonnier and I released a new preprint focusing on the problem of path inference: given a path prefix, i.e., a partially observed sequence of nodes in a graph, we aim to predict which nodes are in the missing suffix.
For more information, see this blog-post.
Review article on sampling techniques
Nicolas Tremblay and I recently dived into the very interesting literature of sampling methods for spectral clustering. Here is the (pre-print) of our labors:
Approximating Spectral Clustering via Sampling: a Review (under consideration as a chapter of the book “Sampling methods for machine learning“)
In it, we discuss a lot of cool techniques from numerical linear algebra and machine learning (such as Nystrom approximation, Random Fourier Features, Coresets, Landmarks, Sparsification), in addition to summarizing our recent discoveries with random graph filters and coarsening.
New work on graph coarsening
I have been thinking for some time now on how can we reduce the size of a graph without significantly altering its basic properties. In my latest work, I present coarsening algorithms that preserve the spectrum of a graph as well as its community structure.
A demonstration can be found in this blog-post.