Dynamic Network Selection using Kernels

We present a new algorithm for vertical handover and dynamic network selection, based on a combination of multi- attribute utility theory, kernel learning and stochastic gradient descent. We show that this new method is able to improve network selection in a non-stationary mobile environment. Furthermore, since the kernel employed is based on the utility functions for attributes such as Availability, Quality and Cost, the kernel regression in fact gives interpretable results. We present simulation results that demonstrate our algorithm being able to dynamically learn utilities and efficiently select networks.