Turing, Learning and Meta-Reasoning¶
Stephen H. Muggleton
RAE/MSR Chair Professor
Imperial College,UK
Abstract: In his seminal 1950 paper Alan Turing introduced not only the first description of Machine Intelligence to appear in a major academic journal, but a route map of how to achieve it. Turing saw Machine Learning as intrinsic to Machine Intelligence. Although intelligence could be argued to be related to Biology, Physiology and Engineering, Turing's argument can be more easily viewed not as an explanation of human intelligence, but rather a study of the phenomenon of mind within the physical universe. He draws on mathematically abstract notions such as memory capacity to show the necessity for learning in order to achieve intelligence. However, he also argues the necessity of meta-reasoning to achieve effective and powerful forms of learning. Turing's argument endorses the use of logic in this context. In this talk we will show how recent developments in meta-interpretive learning are contributing to achieve Turing's suggestions. The meta-interpretive learning framework allows for smooth, efficient and close integration of previously diverse forms of reasoning, including deduction, abduction and induction through the use of mechanisms which support level changes between first, second and higher-order logics. The talk will show that these mechanisms are intrinsic to the logical description of a Universal Turing machine. Initial implementations of Meta-interpretive Learning are not only efficient, but applicable to hard problems involving predicate invention and recursion. Early applications of this approach will be described in tasks as diverse as learning formal grammars, learning robot strategies and construction of complex pattern recognisers from camera data.
Bio: Prof. Stephen H. Muggleton is Director of Modelling at the Centre for Integrative Systems Biology at Imperial College and holds a Royal Academy of Engineering/Microsoft Research Chair. He received his BSc in Computer Science at the University of Edinburgh in 1982. His PhD research, on the topic Inductive Acquisition of Expert Knowledge was carried out at Edinburgh University under the supervision of Prof. Donald Michie. He was awarded his PhD in 1986. During the period 1986-1990 he was a Turing Institute Research Fellow. In 1990 he was awarded a British SERC Postdoctoral Fellowship. In 1993 he was awarded a 5-year EPSRC Adanced Research Fellowship at Oxford University Computing Laboratory, where he founded and headed the Machine Learning Research Group. During the same year he took up an invitation to the Fujitsu Chair as Visiting Associate Professor at the University of Tokyo. In 1997 he was made Reader by the Distinctions Committee of the University of Oxford and took up the Chair of Machine Learning at the University of York. In July 2001 he took up the Joint Research Council Funded Chair of Computational Inference and Bioinformatics at the Department of Computing, Imperial College. He was elected a Fellow of the American Association for Artificial Intelligence in 2002. In 2005 he became Director of Modelling at the new Centre for Integrative Systems Biology at Imperial College (CISBIC) and in 2008 was elected as both a Fellow of the Institution of Engineering and Technology (IET) and a Fellow of the British Computer Society (BCS). In 2010 he was elected a Fellow of the Royal Academy of Engineering. Stephen Muggleton's intellectual contributions within Machine Learning include the introduction of definitions for Inductive Logic Programming (ILP), Predicate Invention, Inverse Resolution, Closed World Specialisation, Predicate Utility, Layered Learning, U-learnability, Self-saturation and Stochastic logic programs.