Summary:
In the current study, we consider the problem of adaptive distributive learning
in wireless sensor networks. Generally speaking, in adaptive learning, the goal
is to study and develop algorithms, which are capable
of learning through input/output measurements, and adapt in possible changes of
the environment and/or
the unknown system to be estimated. Such techniques play a central role in a
number of applications, as for example in echo cancellers, mobile
communications just to name a few. In the literature, a number of adaptive
algorithms, which are of low complexity and/or converge to a good estimate of
the unknown parameter relatively fast, has been proposed. Moreover, adaptive
techniques, which are capable of tracking changes of unknown system, have been
studied. Nevertheless, it was only recently when the problem of adaptive
distributed learning. In a nutshell, in adaptive distributed learning, we
consider a network consisted of nodes, and each node has access to a number of
measurements. The goal is to estimate an unknown, yet common parameter of
interest. There are, mainly, two issues to be addressed. The first one is how
the sensed measurements will be imposed in the problem and the second one is
how each node will exploit the measurements received from the nodes with which
communication is possible.
In the current study, we will present celebrated algorithms, proposed in the
literature, and we will demonstrate how the previously mentioned schemes are
generalized in the distributed scenario. Moreover, an algorithm that attacks
the problem of malfunctioning nodes and a scheme which reduces the number of
information transmitted, will be studied. Finally, the previous techniques will
be experimentally verified.
Keywords:
Adaptive learning, Distributed learning, Wireless sensor networks, Wiener filter, Convex Analysis
