The Economic Foundation for Neural Computing Systems

点击翻译按钮获取中文摘要

This is a Small Grant for Exploratory Research which uses the framework of mathematical economics to model the coordination of amultitude of computational units engaged in large-scale problem solving. The authors are working on developing a detailed formalism of a massively parallel computing system that lends itself to a representation as an economic resource allocation problem. A computation will be regarded as a production process that yields a specific output and requires certain inputs. Just as production of commodities are organized into parallel and serial stages with intermediate goods, a computational problem will be organized into parallel and serial steps with intermediate outputs. Inputs and outputs will be priced, and profits will be defined for each computational unit. Entry and exit of individual units based on rofitability can be interpreted as a form of adaptive learning. Mathematical economic will allow the characterization of optimal prices that will achieve the system objective efficiently. The authors will study the existence, optimality, and stability of equilibrium prices and allocations, the practical implementation of this economic solution, and the potential learning behavior driven by profitability. The research is significant in that it explores the way in which knowledge from economic science and knowledge fromcomputer science can be integrated to produce new knowledge.