A Self-organizing Multi-agent System for Adaptive Continuous Unsupervised Learning in Complex Uncertain Environments

Introduction. Continuous learning and online decisionmaking in complex dynamic environments under conditions of uncertainty and limited computational recourses represent one of the most challenging problems for developing robust intelligent systems. The existing task of unsupervised clustering in statistical learning requires the maximizing (or minimizing) of a certain similarity-based objective function defining an optimal segmentation of the input data set into clusters, which is an NP-hard optimization problem for general metric spaces and is computationally intractable for real-world problems of practical interest (Davidson and Ravi 2005). The task of continuous online learning in complex dynamic environments assumes near real-time mining of streaming data continually arriving at the system, which imposes additional requirements for continuous clustering algorithms. This paper describes the developed computationally efficient adaptive multi-agent approach to continuous online clustering of streaming data in complex uncertain environments and a knowledge-based self-organizing multi-agent system for implementing it. A multi-agent approach to adaptive continuous learning. We address the problem of continuous online learning in changing environments by developing a hybrid learning approach to be both intelligible and computationally efficient that combines multiagent distributed resource allocation and model-based reinforcement learning of POMDPs. The developed multi-agent approach to adaptive online unsupervised learning of streaming data is based on an asynchronous message-passing method of continuous agglomerative hierarchical clustering and a knowledge-based competitive multi-agent system for implementing it. The proposed computationally efficient multi-agent algorithm for online agglomerative hierarchical clustering of streaming data is different from conventional unsupervised learning methods by being distributed, dynamic, and continuous. Distributed clustering process provides the ability to perform efficient run-time learning from both centralized and decentralized data sources without an additional centralized algorithm of aggregating partial mining results. Both the input dataset of decentralized sources and decision criteria for learning (e.g. similarity matrices and expert knowledge) are not fixed and can be changed at runtime during execution of the dynamic algorithm. The con-