Notification-Oriented and Rete Network Inference: A Comparative Study

摘要

This paper presents a practical and a theoretical comparative study between the so-called Notification-Oriented Inference (NOI) and Rete Network inference. NOI is composed of a network of small computational entities, which communicate only when necessary, avoiding both temporal and structural redundancies in a program inference. The innovative aspect of the considered approach would be that computational entities carry out the inference by means of collaborations based upon direct notifications (reactive inference). This approach differs from usual inference mechanisms of rule-based systems, which are based on checks/searches over passive data. This search based execution is present even on advanced mechanisms such as Rete. In this sense, the focus of this paper is to analyze performance measurements over an experimental application implemented in both mechanisms. Coupling and redundancy problems are also considered in a theoretical analysis. The results show that NOI can perform around 6 times faster than the compiled Rete mechanism and around 20 times faster than the interpreted Rete mechanism in the cases proposed in this paper. Moreover, NOI presents better asymptotic complexity time and decoupling degree, presenting itself as a promising alternative to compose complex, distributed and scalable applications.