Learning problem-solving concepts by reflecting on problem solving

Abstract

Abstract. Learning and problem solving are intimately related: problem solving determines the knowledge requirements of the reasoner which learning must fulfill, and learning enables improved problem-solving performance. Different models of problem solving, however, recognize different knowledge needs, and, as a result, set up different learning tasks. Some recent models analyze problem solving in terms of generic tasks, methods, and subtasks. These models require the learning of problemsolving concepts such as new tasks and new task decompositions. We view reflection as a core process for learning these problem-solving concepts. In this paper, we identify the learning issues raised by the taskstructure framework of problem solving. We view the problem solver as an abstract device~ and represent how it works in terms of a structurebehavior-function model which specifies how the knowledge and reasoning of the problem solver results in the accomplishment of its tasks. We describe how this model enables reflection, and how model-based reflection enables the reasoner to adapt its task structure to produce solutions of better quality. The Autognostic system illustrates this reflection process.

Adaptation-guided retrieval: questioning the similarity assumption in reasoning

“…One of the major assumptions in Artificial Intelligence is that similar experiences can guide future reasoning, problem solving and learning; what we will call, the similarity assumption. The similarity assumption is used in problem solving and reasoning systems when target problems are dealt with by resorting to a previous situation with common conceptual features (see e.g., Aamodt & Plaza, 1994;Aha et al, 1991;Carbonell, 1986;Kambhampati & Hendler, 1992;Quinlan 1979Quinlan , 1986Stroulia & Goel, 1994;Winston, 1980). In machine learning, such common features are grist to the mill of inductive learners and concept classifiers based on the assumption that situations with shared features reflect critical distinctions between different classes of situation (see e.g., Cheeseman et al, 1988;Domingos, 1995;Hunt, 1966;Porter et al, 1990;Quinlan, 1986;Stanfill & Waltz, 1986).…”

Few-Shot Induction of Generalized Logical Concepts via Human Guidance

“…Our approach to Concept Learning is closely related to Stroulia and Goel ( 1994 )'s work, which learns logical problem-solving concepts by reflection. G oci 's scoring metric is more general and applicable to both concepts and plans and can be used for learning from a few examples.…”

“…Subsequently, using this concept, the agent can learn to build a rectangular base (with 2 L-shapes ) from another single demonstration and so on till the tower is fully built. Concept learning has been considered as problem solving by reflection (Stroulia and Goel, 1994 ), mechanical compositional concepts (Wilson and Latombe, 1994 ), learning probabilistic programs (Lake et al, 2015 ), etc. While successful, they are not considered in one-shot learning except with SVM (Tax, 2001 ), or with a neural network (Kozerawski and Turk, 2018 ).…”