The AIRC currently has a fully funded PhD position in machine learning. Details are given below and candidates interested in making an application should visit:
Project Description: Project Description: Inductive-machine-learning-based automated fraud detection systems are best treated as a regression problem, and rely on large collections of historical data labelled with known outcomes for training. Adding known outcomes to such collections can be an expensive process, but the use of active learning can alleviate this problem. The key issue in active learning research is the design of the selection strategies used to select only the most informative examples from a larger collection for expert review. Active learning research primarily focuses on classification problems and there remain opportunities to improve selection strategies for regression problems.
The first part of this project will investigate novel hybrid selection strategies for active learning for regression problems. This work will leverage existing work focused on the development of active learning selection strategies based on intrinsic properties of a dataset, and the use of prediction model outputs for novelty detection. The second part of the project will investigate the use of visualisation techniques to allow analysts guide the active learning process using heir own insights. This work will build on existing work on using visualisation for interactive machine learning problems, of which active learning is one example.
Benefits: The scholarship includes a student stipend of €16,000 per annum for four years and all registration fees covered for the duration of the scholarship.
Student Requirements: Minimum of 2.1 in BSc in Computer Science or similar.
Supervisor: Dr Brian Mac Namee
Deadline to Receive Applications: Friday, 01st November 2013.