SC²S Colloquium - June 9, 2010
|Time:||15:00 pm, s.t.|
Narek Melik-Barkhudarov: Spatially Adaptive Semi-Supervised Learning (MA)
The conventional approach to classification problems relies on analyzing the labeled (training) data to construct a classifier. In many applications however unlabeled data is available together with labeled data and its distribution can be used to determine the class separation manifold. This is achieved by means of additional regularization with regards to the intrinsic geometry of the data distribution.
This approach is known as semi-supervised learning as opposed to supervised learning which deals with labeled data only. The aim of this Master's project is to conduct semi-supervised learning with discretization on sparse grids. Higher dimensional problems are aimed for using adaptivity.