NeTS: Small: Machine Learning Meets Wireless Network Optimization: Exploring the Latent Knowledge

Illinois Institute Of Technology, Chicago IL

Investigators

Abstract

Machine learning has been widely applied in various areas including wireless networking. While the capability of machine learning in classification and pattern recognition has been widely accepted, the role it can play on fundamental research issues in wireless networks is yet to be explored. With the proliferation of heterogeneous networking, wireless network optimization has seen a tremendous increase in problem size and complexity, calling for a paradigm of efficient computation. This project aims at a pioneering study on how to exploit deep learning for significant performance gain in wireless network optimization. Innovative techniques are to be developed for extracting latent knowledge from historical optimization instances, and such knowledge will be leveraged to greatly mitigate the computation complexity in solving new optimization problems. The proposed research seamlessly integrates studies in the areas of optimization, machine learning, graph theory, and wireless networking. This interdisciplinary research will not only provide various training projects to undergraduate and graduate students, but also inspire students to pursue high-quality research with an open-minded and cross-disciplinary perspective. Outcomes from this project may directly benefit the industry with low-complexity yet efficient resource allocation algorithms in practical wireless networks. This project is expected to contribute a series of new insights and innovative methods in integrating machine learning with wireless network optimization. This study will reveal that properly trained machine learning algorithms can smartly identify critical features (in terms of a small set of critical links or transmission patterns) that lead to optimal or close-to-optimal solutions. The traditional learning framework for data classification cannot be easily tailored for exposing the latent knowledge in wireless network optimization. This project will conduct a systematic study including learning method selection, input/output design, cost function design, training set construction, and parameter tuning, to accommodate the unique needs and requirements for learning from historical optimization instances. This study will demonstrate how the learned knowledge can be exploited to significantly mitigate the computation cost in both centralized optimization and online scheduling. This study will enable people, possibly for the first time, to understand the complex relationship among the input data traffic, internal network features (link or pattern activation), and optimal resource allocation (scheduling or routing). This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.