Meta-Interpretive Learning and Program Induction

Опубликовано 27 июня 2016, 18:39

This talk will review work at Imperial College on the development of Meta-Interpretive Learning (MIL), a technique which supports efficient predicate invention and learning of recursive logic programs by way of abduction with respect to a meta-interpreter. The approach has been applied to the learning of regular and context-free grammars, and further extended to learn dyadic datalog programs. An extension of the approach uses a meta-interpreter of Stochastic Logic Programs (SLP) to implement a Bayesian posterior distribution over the hypothesis space. An ongoing application of MIL technology will be described in which MIL technology is applied to incrementally learn a series of string transformation program induction problems previously studied by Sumit Gulwani (Microsoft Redmond). In this case learning is constrained to the provision of a small number of examples supplied by a spreadsheet user.