Meta-interpretive learning as metarule specialisation

Abstract In Meta-interpretive learning (MIL) the metarules, second-order datalog clauses acting as inductive bias, are manually defined by user. this work we show that metarules for MIL can be learned MIL. We define a generality ordering of $$\theta$$ θ -subsumption and user-defined sort derivable specialisation most-general matrix in language class; these turn third-order punch with variables quantified over set atoms which only an upper bound on their number literals need user-defined. cardinality metarule is polynomial metarules. re-frame resolution. modify operator to return new rather than first-order prove correctness operator. implement TOIL, sub-system system Louise. Our experiments progressively replaced Louise’s predictive accuracy training times maintained. conclude automatically derived replace