Integration of supervised and unsupervised learning
In machine learning for classification problems, there are two
distinct approaches to learning or classifying data: the
supervised learning and un-supervised learning. The supervised
learning deals with problem where a set of data are labeled for
training and another set of data would be used for testing. The
un-supervised learning deals with problem where none of the labels
of the data are available.
In recent years, important classification tasks have emerged with
enormous volume of data. The labeling of a significant portions of
the data for training is either infeasible or impossible.
Sufficient labeled data for training are often unavailable in data
mining, text categorization and web page classification.
Kernel Methods and Graph Networks for Machine Learning
Kernel methods and graph networks are powerful approaches for solving various machine learning problems.
In particular, randomized methods for optimizing criterion function on graph networks
are interesting solutions to various NP-hard learning problems.
Demonstration for the Randomized Contraction methods for Machine Learning.