Abstract not found

The theory-based model of categorization posits that concepts are represented as theories rather than as feature lists. Thus, it is particularly interesting that the DSM-IV (American Psychiatric Association, 1994), establishes a set of atheoretical guidelines for diagnosis in the domain of mental disorders. Five experiments investigated how clinicians handle an atheoretical nosology. Clinical psychologists' causal theories for DSM-IV disorders and their responses on diagnostic and memory tasks were measured. Participants were more likely to diagnose a hypothetical patient with a disorder if that patient had causally central rather than causally peripheral symptoms according to their theory of the disorder. They also showed biased memory for the causally central symptoms. Clinicians are cognitively driven to form and apply theories despite decades of training and practice with the DSM's atheoretical guidelines. Clinical Psychologists' Theory-Based Representations of Mental Disorders Predict their Diagnostic Reasoning and Memory The theory-based view of categorization proposes that concepts are represented as theories or causal explanations. Murphy and Medin (1985) suggested that our nave theories about the world hold the features of a concept together in a cohesive package. For instance, a layperson's concept of anorexia not only contains the features "fear of becoming fat" and "refuses to maintain minimal body weight," but also the notion that the fear of becoming fat helps cause the refusal to maintain minimal body weight (Kim & Ahn, 2002). Indeed, a growing body of evidence supports the notion that the human mind constantly seeks out rules and explanations that make sense of incoming data concerning its surroundings, and forms concepts based on its theories about the ...

This thesis demonstrates how the power of symbolic processing can be exploited in the learning of low level control functions. It proposes a novel hybrid architecture with a tight coupling between a variant of symbolic planning and reinforcement learning. This architecture combines the strengths of the function approximation of subsymbolic learning with the more abstract compositional nature of symbolic learning. The former is able to represent mappings of world states to actions in an accurate way. The latter allows a more rapid solution to problems by exploiting structure within the domain. A control function is learnt over time through interaction with the world. Symbols are attached to features in the functions. The symbolic attachments act as anchor points used to transform the function of a previously learnt task to that of a new task. The solution of more complex tasks is achieved through composing simpler functions, using the symbolic attachments to determine the composition. T...

Using formal specifications to represent software components facilitates the determination of reusability because they more precisely characterize the functionality of the software, and the well-defined syntax makes processing amenable to automation. This paper presents an approach, based on formal methods, to the modification of reusable software components. From a two-tiered hierarchy of reusable software components, the candidate components that are analogous to the query specification are retrieved from the hierarchy. A retrieved component is compared to the query specification to determine what changes needed to be applied to the corresponding program component in order to make it satisfy the query specification. 1 Reuse and Analogical Reasoning The major objectives of a reuse system are to classify the reusable components, to retrieve them from an existing library, and to modify the retrieved components to satisfy the query specification. In previous investigations, the proces...

Most of the work on how people learn from examples in a new domain has tended to be within the framework of analogical problem solving (APS). APS involves retrieving an analogue from long-term memory and adapting it to fit the current problem. In this paper I argue that this view is inappropriate, and that novices imitate examples to solve new problems. This model does not assume that novices have a useful representation of an earlier problem in memory, nor that they can manipulate a solution, even when one is presented to them. Instead, they tend to see the similarity between problems in terms of their surface features. Using an analogue involves attempting to perform the same operations in the current problem as were performed in the source in a step by step fashion. Is analogical problem solving always analogical? The case for imitation. Second draft 3 1. Introduction Very often the simplest way of solving a problem is to think of a similar one we have solved in the past and use ...

Learning and reasoning are both aspects of what is considered to be intelligence. Their studies within AI have been separated historically, learning being the topic of neural networks and machine learning, and reasoning falling under classical (or symbolic) AI. However, learning and reasoning share many interdependencies, and the integration of the two may lead to more powerful models. This dissertation examines some of these interdependencies and describes several models, culminating with a system called FLARE (Framework for Learning And REasoning). The proposed models integrate inductive learning with prior knowledge and reasoning. Learning is incremental, prior knowledge is given by a teacher or deductively obtained by instantiating commonsense knowledge, and reasoning is non-monotonic. Simulation results on several datasets and classical commonsense protocols demonstrate promise.

Abstract not found

Transfer learning allows knowledge to be extracted from auxiliary domains and be used to enhance learning in a target domain. For transfer learning to be successful, it is critical to find the similarity between auxiliary and target domains, even when such mappings are not obvious. In this paper, we present a novel algorithm for finding the structural similarity between t-wo domains, to enable transfer learning at a structured knowledge level. In particular, we address the problem of how to learn a non-trivial structural similarity mapping between two different domains when they are completely different on the representation level. This problem is challenging because we cannot directly compare features across domains. Our algorithm extracts the structural features within each domain and then maps the features into the Reproducing Kernel Hilbert S-pace (RKHS), such that the “structural dependencies” of features across domains can be estimated by kernel matrices of the features within each domain. By treating the analogues from both domains as equivalent, we can transfer knowledge to achieve a better understanding of the domains and improved performance for learning. We validate our approach on a large number of transfer learning scenarios constructed from a real world dataset.