An active and incremental learning framework for the online prediction of link quality in robot networks. (January 2019)

• Record Type: Journal Article
• Authors: Lowrance, Christopher J.; Lauf, Adrian P.
• Is Part Of: Engineering applications of artificial intelligence. Volume 77(2019)
• Page Start: 197
• Page End: 211
• Publication Date: 2019-01
• Subjects: Single-pass active learning in data streams -- Incremental machine learning with forgetting -- Concept drift and cold-start mitigation; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2018.10.006
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: This paper presents a comprehensive framework for the active and incremental learning of link quality (LQ) in robot networks. Mobile robots need foresight into the quality of their wireless links in order to proactively optimize routing, plan mobility routes, avoid disconnects, and make other network optimizations. However, the task of predicting LQ is nontrivial. Many environmental factors that influence the quality of radio wave propagation are dynamic, and thus, robots must continually learn and update their prediction models while they operate online. Prior work in the field uses online learning algorithms for predicting LQ, but this approach is costly in terms of energy and network capacity because of the need for a consistent stream of LQ labels to be transmitted from the receiver to the transmitter. Hence, this paper introduces a framework to reduce these overhead expenses by incorporating active learning to selectively label only a portion of the samples from the data stream. The framework also uses incremental training batches to conserve labeling resources, and updates the batches using change detection and forgetting mechanisms to mitigate concept drift. Experimental results reveal that the framework reduces label queries by up to 21.5% and prediction error by up to 9% after periods of concept drift.