Cooperative heterogeneous intelligent processing systems: Applications and tools.

摘要

The human brain may be described as a group of components, physically distinct and separated but in full communication. This description, applicable to computing systems as well, is a motivator for simulating brain-like computation on a computer. This dissertation thus takes neural network models as a prime focus.;Most artificial neural network (ANN) models discussed are modular in nature and adaptable to heterogeneity across modules. This thesis, however, extends this heterogeneity to include conventional algorithm modules, which provides the rationale for the characterizing phrase, cooperative heterogeneous intelligent processing system(s) (CHIPS).;CHIPS cooperate to the extent that their heterogeneous elements work together to accomplish tasks of an intelligent processing system; they exploit forms of computation in a manner suitable to the needs of the task. The Introduction develops this CHIPS notion in the context of problems to be solved, ANN models relied upon (e.g., fast learning nets), and tool development (given that models must be implemented).;Beyond contributing to the development of modular neural network and CHIPS computational styles, this research's predication of implemented models has led us to significant advances in tools for building models and analyzing their results, in the spirit of the BEAK (Build Execute Analyze Knowledge) model of the thesis' advisor (Reilly, Barrett, Tarng, & Hyatt, 1995). Tool exploration has settled upon an integrated collection of methods that include a CHIPS connectivity and communication library and command line interface, Stuple Space, an ANN-oriented vector library, vecmat, and a set of statistical programs and a graphical display program, for monitoring and analyzing program output. The resulting set of code forms a CHIPS development kit.;The collection of increasingly comprehensive artificial intelligence (AI) models in this document, with major applications to human behavior, include as a central problem of study the use of strategies by children. The CHIPS kit matures over three phases, in step with the increasing sophistication of the applications. Several published articles document our efforts through these phases; the reader is directed toward the appendices for these. Advancement of these models and the above mentioned tools remains an ongoing effort.