Swansea University, UK
Pavel Loskot is a Senior Lecturer in System and Information Engineering at Swansea University, UK. He has an extensive portfolio of industrial and academic collaborative projects involving both small and large companies and institutions. He has been involved in telecommunication research and development since 1996 working across the whole protocol stack. His current research is concerned with problems in signal processing, tactical mobile networks, Internet access in challenging environments, and computational molecular biology. He is also the Senior Member of the IEEE, Fellow of the Higher Education Academy in the UK, and the Recognized Research Supervisor of the UK Council for Graduate Education.
Title and abstract
Quest for machine intelligence: From estimation to statistics and machine learning
The enormous growth in connectivity, data, storage, and computing over the past decade drives the need for intelligent information processing systems. Machine learning is particularly attractive for its universality, low number of requirements and assumptions, and relatively straightforward implementation within some ready-made computing frameworks. However, today’s trend of overusing machine learning to solve all problems is somewhat unfortunate. There are many other well understood and verifiable methods of statistical signal processing which likely provide better performance in many scenarios. In this talk, I will explain the differences in approaching problems as statistical estimation tasks, how they are connected to classical methods of statistics, and when machine learning strategies can or should be used. The key factors to take into account is what is known, assumed and expected as well as the availability of relevant data and tractability of mathematical or computational models considered. In addition, there can be a number of other constraining factors such as the maximum algorithmic complexity, off-line or online applications, the scale of deployment, and the limited time for obtaining a working solution. Finally, common tasks and services performed in a typical multi-user communication system will be reviewed to understand where the intelligent decisions can be achieved by the means of classical detection and estimation methods, and where machine learning could bring substantial benefits.