Graphical Semi-Supervised Learning
Probit and regression models on various manifolds
We investigate graphical semi-supervised learning approaches and benchmark both probit and regression models on various manifolds. We construct data on well-separated clusters in the plane, weakly connected clusters on a spiral manifold, and examine data from the MNIST data set. We find that these approaches are generally effective, but can be limited by the underlying data and choices made in model construction.
You can find a detailed report, results, and code here.