Projects
Research ProjectsMembers of the Machine Learning Group are currently actively involved in the following research projects:
Sentiment AnalysisThe MLG is conducting an experiment to gauge sentiment towards Ireland's economy in the Irish media. For this experiment we have developed a system that uses active machine learning techniques to dynamically monitor the balance between positive and negative economic news published in popular online Irish news sources. Clique: Graph & Network Analysis Cluster Clique is a Science Foundation Ireland funded (SFI) Strategic Research Cluster focussed on graph and network analysis and visualisation. Our industry partners, in this joint initiative are IBM, Idiro Technologies and Norkom Technologies. Our academic partners are UCD and DERI at NUI Galway along with collaborating academics around the world. Data Mining - Between Unsupervised and Supervised Techniques (DaM-BUST)This project focuses on real-life applications for machine learning that do not fit neatly into the traditional distinction between supervised and unsupervised techniques. This research specifically involves the analyis of data from different application areas including Bioinformatics (genome & proteomics) and Multimedia (image & text). One-Sided Classification and Inductive Learning (OSCAIL) In traditional supervised classification problems, discriminating classifiers are trained using positive and negative examples. However, for a significant number of practical problems, counter-examples are either rare, entirely unavailable or statistically unrepresentative. Such problems include industrial process control, text classification and analysis of chemical spectra. One-sided classifiers are emerging as a solution, which characterize the target class, to distinguish it from all other classes. Existing commercial machine learning products do not include this family of techniques. In this project we conduct research into new techniques for one-sided classification and produce a comprehensive toolkit for building one sided-classifiers, that can be incorporated into third-party products
|
