Issue No. 06 - December (1991 vol. 6)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/64.108949
<p>A description is given of the FOIL (First-Order Inductive Learner) system, which exploits information from large numbers of examples to guide the search for a program. FOIL develops first-order rules from structured data described by a collection of relations. The guidance it provides turns out to be so effective that greedy search is usually adequate. Algorithms using greedy search, however, tend to suffer from a horizon effect: an action that might be desirable or even essential from a global perspective can appear relatively unpromising at a local level and so may be passed over. Rather than restricting the search space, and thus the class of learnable programs, FOIL exploits determinism to overcome some of the horizon effect of greedy search. The effect on learning time is usually negligible.</p>
J. R. Quinlan, "Knowledge Acquisition from Structured Data: Using Determinate Literals to Assist Search," in IEEE Intelligent Systems, vol. 6, no. , pp. 32-37, 1991.