Networks of classifiers can offer improved accuracy and scalability over single classifiers by utilizing distributed processingresources and analytics. However, they also pose a unique combination of challenges. First, classifiers may be located acrossdifferent sites that are willing to cooperate to provide services, but are unwilling to reveal proprietary information about theiranalytics, or are unable to exchange their analytics due to the high transmission overheads involved. Furthermore, processingof voluminous stream data across sites often requires load shedding approaches, which can lead to suboptimal classificationperformance. Finally, real stream mining systems often exhibit dynamic behavior and thus necessitate frequent reconfiguration ofclassifier elements to ensure acceptable end-to-end performance and delay under resource constraints. Under such informationalconstraints, resource constraints, and unpredictable dynamics, utilizing a single, fixed algorithm for reconfiguring classifiers canoften lead to poor performance. In this paper, we propose a new optimization framework aimed at developing rules for choosingalgorithms to reconfigure the classifier system under such conditions. We provide an adaptive, Markov model-based solution forlearning the optimal rule when stream dynamics are initially unknown. Furthermore, we discuss how rules can be decomposedacross multiple sites and propose a method for evolving new rules from a set of existing rules. Simulation results are presented fora speech classification system to highlight the advantages of using the rules-based framework to cope with stream dynamics.
展开▼
