The Department of Computer Science at the University of Cyprus cordially invites you to the PhD Defense entitled:
Integrating Temporal Abstraction with Bayesian Networks: A Validation in the field of Coronary Heart Disease
Speaker: Kalia Orphanou
This thesis explores the integration of two Artificial Intelligence technologies: Temporal abstraction (TA) and Bayesian networks (BN) in order to improve medical problem solving. In clinical systems, given the inherent uncertainty and incompleteness of medical knowledge and data, probabilistic models became a popular representation for reasoning with disease processes. Bayesian networks belong to the family of such probabilistic models and in fact they were widely used in many clinical domains as they can handle well uncertainty in medical knowledge and data. The time-stamped multivariate data representing the medical history of some individual are usually not amenable to direct reasoning. Temporal abstraction (TA) methods, by combining statistical and heuristic methods aim to glean out the useful information/patterns from such data, in order to facilitate specific uses including higher level problem-solving. The generated, more abstract (and hence more useful/usable) information is of different types that can be roughly divided into basic and, complex temporal abstractions. Bayesian networks and Temporal abstraction, demonstrated their effectiveness as standalone engines, predominantly for medical problem solving, but not in conjunction. The key research hypothesis that this thesis set out to investigate was whether the integration of TA with BN could yield notable performance improvements in medical problem solving. Towards this end, we selected the field of coronary heart disease (CHD) as a testbed and demonstrator of the attempted integration. Overall the novel contributions of this thesis are: a) the development of a temporal extension of a BN, namely a Dynamic Bayesian network, whose nodes represent basic TAs and its application for diagnosis and prognosis (primary and secondary prevention) of CHD, b) the development of a Naive Bayes classifier whose features represent frequent temporal association rules (TARs), a type of complex TAs, applied for the diagnosis of CHD and c) the formulation of a general methodology for the proposed integration, potentially applicable to any domain where time is important.
Kalia Orphanou is a PhD candidate in the University of Cyprus. Her research interests include: Bayesian networks, temporal abstraction, temporal reasoning, temporal data mining and their applications in medical expert systems. She has an MEng in Computer Science with Artificial Intelligence degree from the University of Southampton (2009). During her doctoral studies, she worked as a Teaching Assistant at the University of Cyprus from 2010 to 2013 and she also joined the Laboratory for Internet Computing (LINC) from 2013 to 2015. She has participated in the Erasmus+ mobility program during the period of 01/10/2015 – 29/02/2016 where she was hosted in the Laboratory of Biomedical Informatics at the University of Pavia in Italy. Recently, she has been awarded an ERCIM Alain Bensoussan Fellowship Programme, and will be hosted by the Centrum Wiskunde & Informatica (CWI), in Netherlands, for one year (April 2017 – March 2018).
|Other Presentations Web: https://www.cs.ucy.ac.cy/colloquium/presentations.php|
|Colloquia Web: https://www.cs.ucy.ac.cy/colloquium/|