Network management and configuration is an essential attribute of any wireless network with reliable self-tuning capabilities. However,the cost and overhead of network management has rarely been accounted for from a fundamental limit (information theoretic) perspective. In contrast to the past generations of networking solutions, on the other hand, in the ever-increasingly mobile and large-scale networks of tomorrow the network reconfiguration overhead may not be insignificant; this includes the initial beam alignment, link maintenance, spectrum sensing, packet resizing, etc. Our work aims to provide fundamental limits on the overhead associated with learning, network tuning, and optimization of network parameters.
Our approach relies on fundamental notions in information theory and statistics to quantify the networking overhead and utilizes recent data analytic and machine learning algorithms to develop practical learning/optimization algorithms. In the first part of the talk, we consider the problem of reliably and quickly searching for a parameter of interest in a large signal space in face of measurement-dependent noise. This problem naturally arises in many practical communications systems such as the directional link establishment and maintenance (beam alignment) as well as spectrum sensing for cognitive radios. In the second part of the talk, we consider an important variant of the search problem: data-driven (Bayesian and non-Bayesian) function maximization and its connection to network parameter tuning.
Sponsored by IEEE Santa Clara Valley Information Theory Society. Please register for this event.