Approximate Reasoning by Agents

Abstract

We present an approach to approximate reasoning by agents in distributed environments based on calculi of information granules. Approximate reasoning schemes are basic schemes of information granule construction. An important property of such schemes is their robustness with respect to input deviations. In distributed environments, such schemes are extended to rough neural networks that transform information granules into information granules rather than vectors of real numbers into (vectors of) real numbers. Problems of learning in rough neural networks from experimental data and background knowledge are outlined.