Improved transductive regression using interconnected data
The explosion of data from personal phones, apps, and sensors have enabled powerful machine learning algorithms to help computers identify, categorize, and evaluate information without the help of humans. However, teaching computers how to identify, categorize, and evaluate information usually requires feeding the computers a lot of data pre-labelled by humans. The pre-labelling process is costly and time consuming. The goal of this project is to develop new algorithms to teach computers to identify, categorize, and evaluate information with less pre-labelled data. These new algorithms will use hidden relationships within the information itself, and will also be able to integrate other related data in a way that is easier for the computer use. If successful, these algorithms will be able to learn more from data, while reducing the burden of producing pre-labelled data.