Design and implementation of hardware for real-time intelligent agents

Abstract

Creation of intelligent agents (IAs) became a basic trend in artificial intelligence in recent years. One of the most important requirements to such systems is real-time decision making when operating in open dynamic and uncertain environments. The efficiency of IAs in hard real-time applications can be improved by hardware support. Rule-based (production) knowledge are widely used in different subsystems of IAs to represent both domain and control knowledge.In this paper a known approaches to design and implementation of inference processor for rule-based systems are considered and a novel one is presented. An approach is based on the concept of universal interpreter for rule-based systems and its implementation on microprogram level. The design tasks and processor's architecture based on Rete-network with multiple input βnodes are considered. A design methodology for such class of devices and corresponding software tool for design process support are described. An implementation of processor using Altera programmable logic devices (PLD) and performance estimation results are presented in conclusion.

IntelliSoC: A System Level Design and Conception of a System- on-a-Chip (SoC) to Cognitive Agents Architecture

“…Still searching for dedicated hardware for the Rete algorithm, [12] present a special purpose machine that is connected to a host computer as a support for agents execution in software. The authors also shown that increasing the number of inputs of the beta nodes in the network generated by the algorithm enhances the performance of the proposed hardware: using four or five inputs in a join node of the beta network allowed for a 15-20% increase in performance.…”