Application Keynote Lecture
Prof. Frans Coenen (University of Liverpool, UK)
The Discovery of Knowledge in Image Data: A Study in Automated Population Estimation Using The Google Static Maps Service
Image mining is an important element of the canon of data mining. Decision making is routinely supported by visual information and the visualisation of data. At the same time our ability to collect visual information is increasing rapidly, partly because of technological advancements and partly (and associated with the first) the increasingly reduced cost of collecting such data. Consequently the demand for utilising image data for the purpose of extracting information (image mining) is increasing in a corresponding manner. A taxonomy for image representation in the context of image mining is thus presented. The main premise being that the actual mining algorithms that may be used are well understood, it is the pre-processing of the image data that remains a challenge. The requirement for the output from this pre-processing is some image representation that is both sufficiently expressive while at the same time being compatible with the mining process to be applied. Three categories of representation are considered: statistics based, tree (graph) based and point series based. The suggested taxonomy is then analysed in further detail by considering a novel image mining application, the collection of individual household census data from Google earth satellite imagery. The representations are considered both in terms of generating census prediction models (using classification and regression) and in terms of applying such models for the purpose of larger scale census prediction.
Frans Coenen has a general background in AI, and has been working in the field of data mining and Knowledge Discovery in Data (KDD) for the last fifteen years. More specifically with respect to the many elements that comprise the research domain of Data Mining and Big Data Analytics as applied to unusual data sets, such as: (i) graphs of all kinds including social network data; (ii) time series such as sound waves for the purposes of identifying patterns and trends; (iii) free text of all kinds, but particularly legal texts, not just for data extraction purposes but also with a view to learning ontologies; (iv) 2D and 3D images, especially medical images and (v) video data. He is also interested in data mining using (homomorphically) encrypted data. He currently leads a small research group (8 PhDs and 3 RAs) working on many aspect of data mining and KDD. He has some 350 refereed publications on KDD and AI related research, and has been on the programme committees for many KDD conferences and related events. He is pleased to have been the founder of the UK KDD symposia series, which is now in its twelfth year. Frans Coenen is a member of The British Computer Society (BCS) and the BCS' Specialist Group in AI (BCS-SGAI). He has been chair and deputy/technical programme chair for the BCS-SGAI AI series of conference. Frans Coenen is currently professor within the Department of Computer Science at the University of Liverpool where he is the director of studies for the department's on-line MSc programmes.