ACCEL is a general purpose system that uses abductive reasoning to construct explanations for observed intelligent phenomena. These explanations are then used to avoid redundant work in future problem solving episodes. We define an abductive explanation as a consistent set of assumptions which when combined with background knowledge, logically entails a set of observations.
ACCEL has been constructed as a domain-independent system, in which knowledge about a variety of domains has been uniformly encoded as first-order Horn-clause axioms. A general-purpose abduction algorithm, AAA, is used to efficiently construct explanations by caching partial explanations. ACCEL has been shown to achieve more than an order of magnitude speedup in run time for a variety of domains, including plan recognition in text understanding, and diagnosis of medical diseases, logic circuits, and dynamic systems.
For more information, see here.
An ILP system that integrates traditional top-down and bottom-up approaches to combine the strengths of each and eliminate the weaknesses of both. For more information, on downloading and using the system, see the Beth Manual.
CHILL (Constructive Heuristics Induction for Language Learning) is a general approach to the problem of inducing natural language parsers. Given a suitably annotated corpus, CHILL produces a parser for mapping subsequent sentences into representations. It treats parser induction as the problem of learning rules to control the actions of a shift-reduce parser expressed as a Prolog program. Control rules are induced by utilizing a novel Inductive Logic Programming (ILP) algorithm, namely CHILLIN, that has been developed to handle the issues arising in the natural-language control-rule domain. For more information, click here.
CHILLIN is an ILP system which integrates top-down search (a la FOIL), bottom-up search (a la GOLEM) and predicate invention (a la CHAMP). It has been used as the induction algorithm underlying the CHILL natural-language parser learning system. For more information and to download the system, click here.
DOLPHIN is a system which combines Inductive Logic Programming (i.e. FOIL) and Explanation-Based Learning (i.e. EBG) to learn search-control rules for eliminating backtracking in Prolog programs. For more information and to download the system, click here.
FOIDL is an ILP system for learning first-order decision lists (ordered lists of clauses each ending in a cut). It has been used to obtain the current best results on learning the past-tense of English. For more information, and to download both Prolog and Common Lisp code, click here.
FORTE (First Order Revision of Theories from Examples) is a machine learning system for modifiying a first-order Horn-clause domain theory to fit a set of training examples. FORTE uses a hill-climbing approach to revise theories. It identifies possible errors in an input theory and calls on a library of operators to develop possible revisions. These operators are constructed from methods such as propositional theory refinement, first-order induction, and inversion resolution. To download and for more information, click here.
NEITHER is a propositional theory refinement system that will modify a incomplete or incorrect rule base so as to make it consistent with a set of input training examples. NEITHER has been extended to revise both Horn clauses and M-of-N rules. An iterative greedy method is used to efficiently compute repairs.
NEITHER has also been used as part of separate system, ASSERT, which performs student modeling. ASSERT (Acquiring Stereotypical Student Errors using Refinement of Theories) is an intelligent tutoring system which inputs a knowledge base describing a domain and a set of student errors on that domain and outputs a tutoring program tailored to fit student needs. Student behavior on the domain is modeled by collecting any refinements to the knowledge base (made by NEITHER) that were necessary to account for the student's behavior. These models are then used to generated feedback which should help raise student performance on that domain. More information can be found here.
RAPIER is a bottom-up inductive learning system for learning information extract rules. It has been tested on several domains and performs comparably to or slightly better than other recent learning system for this task. Code in C++ can be downloaded from here.