Poster Abstract: Link Quality Estimation—A Case Study for On-line Supervised Learning in Wireless Sensor Networks
We focus on the implementation issues of on-line, batch supervised learning in computationally limited devices. As a case study, we consider the use of such techniques for link quality estimation. We compare three strategies for the on-line selection of the data samples to be kept in memory and used for learning. Results suggest that strategies that keep balanced the set of training samples in terms of ranges of target values provide better accuracy and faster convergence.
- 2.Di Caro, G.A., Kudelski, M., Feo, E., Nagi, J., Ahmed, I., Gambardella, L.: On-line supervised learning of link quality estimates in wireless networks. In: Proceedings of the 12th IEEE/IFIP Annual Mediterranean Ad Hoc Networking Workshop, pp. 69–76 (2013)Google Scholar
- 3.Rasmussen, C.E., Williams, C.K.I.: Gaussian Processes for Machine Learning. MIT Press, Cambridge (2006)Google Scholar
- 4.Gibbs, M., MacKay, D.J.: Efficient Implementation of Gaussian processes. Technical report, Cavendish Laboratory, Cambridge (1997)Google Scholar
- 5.Doddavenkatappa, M., Chan, M.C., Ananda, A.: Indriya: A low-cost, 3D wireless sensor network testbed. In: Proceedings of TRIDENTCOM, pp. 302–316 (2011)Google Scholar