Spatio-Temporal Nonlinear Learning with Echo State Wireless Sensor Networks

The proposed research addresses the question of nonlinear time- and space-dependent machine learning that combines recursive neural networks and wireless sensor networks. Within the proposed concept - Echo State Wireless Sensor Networks (ES-WSN), the sensors are equipped with simple computational capabilities, allowing them to perform neuron-like data processing. The resulting sensor network can thus be used to construct models of general nonlinear space- and time-varying processes. This research is aimed at i) studying and extending the learning abilities of ES-WSNs, and ii) designing efficient learning algorithms that exploit ES-WSN network infrastructure for data acquisition and processing. The proposed research not only extends the methodology of machine learning for modeling space- and time-dependent processes, but also offers a number of important practical implications. The ES-WSN and the developed learning algorithms will find their application in the area of cognitive technical systems. Such systems, equipped with numerous sensors, will then be able to learn based on their sensory input data in order to analyze, predict, and react to external excitation in an intelligent way. The results of the proposed research will be distributed in the form of publications and Habilitation thesis.