QuickFOIL

Abstract

Inductive Logic Programming (ILP) is a classic machine learning technique that learns first-order rules from relationalstructured data. However, to-date most ILP systems can only be applied to small datasets (tens of thousands of examples). A long-standing challenge in the field is to scale ILP methods to larger data sets. This paper presents a method called QuickFOIL that addresses this limitation. QuickFOIL employs a new scoring function and a novel pruning strategy that enables the algorithm to find highquality rules. QuickFOIL can also be implemented as an in-RDBMS algorithm. Such an implementation presents a host of query processing and optimization challenges that we address in this paper. Our empirical evaluation shows that QuickFOIL can scale to large datasets consisting of hundreds of millions tuples, and is often more than order of magnitude more efficient than other existing approaches.

Induction of Non-monotonic Logic Programs To Explain Statistical Learning Models

“…In contrast, top-down approach starts with the most general clauses and then specializes them. A top-down algorithm guided by heuristics is better suited for large-scale and/or noisy datasets [22], particularly, because it is scalable.…”

“…The use of a greedy heuristic allows FOIL to run much faster than bottom-up approaches and scale up much better. For instance, the QuickFOIL system [22] can deal with millions of training examples in a reasonable time.…”

Guided Inductive Logic Programming: Cleaning Knowledge Bases with Iterative User Feedback

“…After three to eight iterations, the means of the precision intervals converge to about 0.91 for T + and even to 0.94 for T − , which confirms that our iterative GILP approach is particularly suitable for pruning negative facts. ILP is widely applied in many applications, such as event recognition [13], data cleaning [22], classification [18], and many others. In this paper, we extend ILP to handle a non-fixed training set, while the learning process is guided by user feedback in an iterative and interactive manner.…”

“…There are a number of recent approaches that specifically tackle the problem of learning consistency constraints from a given KB (or, respectively, from a fixed training subset of the KB) for data-cleaning purposes (see [21] for a recent overview). The kinds of constraints considered for data cleaning traditionally comprise functional dependencies, conditional functional dependencies [6], equality-generating dependencies [5], denial constraints and more general Horn clauses [9,22]. The common rationale behind these learning approaches is that "unusual implies incorrect".…”

“…Another major challenge arises due to the fact that the training set of GILP is not fixed but will expand during the learning phase. All recent works in the context of rule learning, such as [9,18,22], take an entire knowledge base as the training set (thus following the assumption that "unusual implies incorrect") and inductively learn rules until a fixpoint condition is reached. In GILP, we have to rely on a possibly small but growing amount of user feedback to both induce and verify the candidate rules, and hence we also aim to discover systematic extraction mistakes that could not be uncovered by traditional ILP approaches.…”