Embedded Files
Peter Lucas, MD PhD, is professor of Artificial Intelligence at the University of Twente. Previously he has been working at various other universities. He has been educated as a medical doctor as well as a computer scientist, and has been involved in research in the area of Artificial Intelligence, in particular knowledge-based systems and Bayesian networks, since the beginning of the 1980s. He has contributed to this area by theoretical as well as applied research, the latter for the major part focusing on the field of clinical medicine. At the moment, his research interests include topics such as applied logic and theorem proving, knowledge representation, decision-support systems, model-based diagnosis, Bayesian networks and statistical machine learning. He is currently involved in a number of multidisciplinary projects aiming to deliver medical decision support systems to oncologists and patients. He has extensively published in AI, computing science and medical informatics journals and conferences with more than 200 papers, wrote and edited a number of AI books, organised a number of workshops in the field, and edited several thematic issues of journals on the topics mentioned above.
Peter Lucas - Is there a place for Medical Knowledge Representation in the Age of Deep Learning?
Peter Lucas - Is there a place for Medical Knowledge Representation in the Age of Deep Learning?
An increasing number of deep learning health-care applications has been developed in recent years, profiting from the availability of successful deep neural network architectures and the potentials of transfer learning. Most of these applications are meant to assist medical doctors in biosignal and image interpretation tasks, but have cast doubts on trustworthiness because of lack of transparency and explainability. In addition, deep learning normally requires much data, whereas the typical clinical, observational dataset is relatively small with much missing values, giving rise to biased results. Hence, methods that complement and integrate data with clinical knowledge from experts or literature are usually of much value and knowledge representation offers the methods to do so. Such methods must be able to deal with uncertainty, one of the essential characteristics of patient data and clinical decision-making. In addition, in some clinical problems there is the extra dimension of time. In the talk we will review some of the features of clinical knowledge, reasoning, learning and clinical data in what way knowledge representation is able to help dealing with clinical problems involving data, and illustrate these by health-care applications developed in my research groups.
Google Sites
Report abuse