Active Learning in Graph Based Semi-Supervised Learning


SIAM CSE 2021, Minisymposium on Theory and Applications of Graph-Based Learning. Presented my work on model change active learning for graph-based semi-supervised learning (SSL), where we use the approximate change in the underlying SSL model as a measure of usefulness in the active learning process. This approximate change is efficiently done for a family of graph-based SSL models, using only a subset of the graph Laplacian’s eigenvalues and eigenvectors.

See my slides here.