@Nguyen Hoang Nghia - Defense date: June 26, 2019

This repository contains my master thesis on semi-supervised learning.

Here we have

• Thesis
• Source code, experimental results: see the last heading below

I examined two generative methods of semi-supervised learning. Here is a quick summary:

• The first is the warm-up part with the common mixture of multinomial models in chapter 2, I argured a trivial problem of the need of initialization of sub-components in class conditional distribution. You may only want to read section 2.3 if you are familiar with this model.
• The main result is in chapter 3 where I employed a simple graphical model for the graph representation of data (graph-based methods). The idea is in section 3.3.

If you want to know more about my motivation, you may refer to section 1.2 of Introduction and the last section of chapter 4.