Machine Learning across the WSN Layers

Abstract

Wireless sensor networks (WSNs) have seen rapid research and industrial development in recent years. Both the costs and size of individual nodes have been constantly decreasing, opening new opportunities for a wide range of applications. Nevertheless, designing software to achieve energy-efficient, robust and flexible data dissemination remains an open problem with many competing solutions. In parallel, researchers have effectively exploited machine learning techniques to achieve efficient solutions in environments with distribution and rapidly fluctuating properties, analogous to WSN domains. Applying machine learning techniques to WSNs inherently has the potential to improve the robustness and flexibility of communications and data processing, while simultaneously optimizing energy expenditure. This chapter concentrates on applications of machine learning at all layers in the WSN network stack. First, it provides a brief background and summary of three of the most commonly used machine learning techniques: reinforcement learning, neural networks and decision trees. Then, it uses example research from the literature to describe current efforts at each level of the stack, and outlines future opportunities.

Introducing DOWSN: Distributed optimization in wireless sensor networks

“…In [24], a fuzzy knowledge-based sensor network is proposed in which each node is able to infer information from its neighbors, thus providing a more accurate and reliable output. An extensive survey of applications of machine learning in WSNs is reported in [25].…”

Distributed optimization in wireless sensor networks: an island-model framework

“…Machine learning represents an alternative means to improve network reliability and prolong network lifetime. An extensive survey of applications of machine learning in WSNs is reported in [13]. For example, in [30] a decentralized Reinforcement Learning (RL) is proposed, where each node is a self-learning agent whose purpose is finding the optimal schedule (i.e.…”

“…This value seems to offer the best trade-off between a deterministic (unitary imitation rate) and an improbable (imitation rate 0.1) acceptance of solution updates: the first condition likely causes an excessive exploitation and a diversity impoverishment, the second excessively promotes exploration and almost suppress, de facto, the effect of the exchange of information. 2.085e + 03 ± 0.00e + 00 2.085e + 03 ± 0.00e + 00 = f 8 1.686e − 05 ± 2.14e − 05 1.583e − 05 ± 7.27e − 06 = f 9 −2.000e + 00 ± 3.80e − 04 −2.000e + 00 ± 4.62e − 05 = f 10 0.000e + 00 ± 0.00e + 00 0.000e + 00 ± 0.00e + 00 = f 11 3.750e − 07 ± 9.92e − 07 3.750e − 07 ± 1.13e − 06 = f 12 0.000e + 00 ± 0.00e + 00 0.000e + 00 ± 0.00e + 00 = f 13 0.000e + 00 ± 0.00e + 00 0.000e + 00 ± 0.00e + 00 = f 14 0.000e + 00 ± 0.00e + 00 0.000e + 00 ± 0.00e + 00 = f 15 4.150e − 06 ± 5.55e − 06 2.250e − 06 ± 2.37e − 06 = 3.580e + 00 ± 2.10e + 00 5.032e + 00 ± 2.71e + 00 = f 6 −8.032e + 00 ± 6.41e − 01 −8.198e + 00 ± 4.56e − 01 = f 7 6.256e + 03 ± 2.75e − 01 6.256e + 03 ± 1.38e − 01f 8 3.902e − 01 ± 1.58e − 01 3.059e − 01 ± 1.08e − 01 = f 9 −1.918e + 00 ± 6.35e − 02 −1.943e + 00 ± 4.73e − 02 = f 10 9.664e − 02 ± 3.81e − 02 7.116e − 02 ± 3.07e − 02f 11 5.152e − 03 ± 1.87e − 03 4.491e − 03 ± 1.17e − 03 = f 12 8.994e − 03 ± 6.54e − 03 7.261e − 03 ± 5.53e − 03 = f 13 3.312e − 02 ± 2.95e − 02 2.941e − 02 ± 1.99e − 02 = f 14 0.000e + 00 ± 0.00e + 00 0.000e + 00 ± 0.00e + 00 = f 15 9.451e − 01 ± 4.44e − 01 7.341e − 01 ± 2.49e − 01 =…”

Genetic Improvement of Routing Protocols for Delay Tolerant Networks

“…In the past two decades, various Machine Learning and bio-inspired technique, especially Evolutionary Algorithms (EAs), have been applied to network problems and particularly protocol optimization. For instance, some researchers have proposed various solutions based on collective intelligence [86] and Reinforcement Learning (RL) [62,72,74] to optimize routing protocols for Wireless Sensor Networks [5,22,37]. Albeit quite powerful, the main limitation of most of these approaches is that they often require a large amount of data collected from the network, to train a model of the protocol, to be used later at runtime for further optimization.…”