Software management oracles often contain numerous subjective features. At each subjective point, a range of behaviors is possible. Stochastic simulation samples a subset of the possible behaviors. After many such stochastic simulations, the TAR2 treatment learner can find control actions that have (usually) the same impact despite the subjectivity of the oracle. 1.

Abstract not found

We draw a simple correspondence between kernel rules and prime implicants. Kernel (minimal) rules play an important role in many induction techniques. Prime implicants were previously used to formally model many other problem domains, including Boolean circuit minimization and such classical AI problems as diagnosis, truth maintenance and circumscription. This correspondence allows computing kernel rules using any of a number of prime implicant generation algorithms. It also leads us to an algorithm in which learning is boosted by an auxiliary domain theory, e.g., a set of rules provided by an expert, or a functional description of a device or system; we discuss this algorithm in the context of SE-tree-based generation of prime implicants. Introduction Rules have always played an important role in Artificial Intelligence (AI). In machine learning, while a variety of other representations have also been used, a great deal of research has focused on rule induction. Moreover, many of the...

Artifical Intelligence (AI) is useful. AI can deliver more functionality for reduced cost. AI should be used more widely but won’t be unless developers can trust adapative, nondeterministic, or complex AI systems. Verification and validation is one method used by software analysts to gain that trust. AI systems have features that make them hard to check using conventional V&V methods. Nevertheless, as we show in this article, there are enough alternative readily-available methods that enable the V&V of AI software.