Empowering Wireless Networks with Collective Wisdom

Artificial intelligence and machine learning technologies are advancing at an accelerating pace nowadays, expected to reshape modern industry and human society in a profound way. The prevalence of a massive number of intelligent devices, supported by high-speed and large-scale connectivity, generates a wealth of data waiting to be explored and utilized. This great opportunity also imposes severe challenges to the current networking system and centralized computing paradigm, as they are not designed to handle the sheer volume and velocity of data flow and make inference and decisions in a timely manner. In addition, many applications are hindered due to privacy concerns involved in data sharing. Federated learning has emerged as a promising solution to address these challenges, and this project aims to advance the frontiers of knowledge and practice of federated learning through exploring the multi-faceted tradeoffs among learning performance, communication efficiency, privacy protection, and system robustness under a generalized hierarchical and hybrid learning framework. The proposed research is expected to facilitate the integration of intelligent services into future wireless systems, provide an attractive business model to promote collaborative exploitation of sensitive data (for example in spectrum sharing and personalized medicine), and help establish a society where data and knowledge can be shared with confidence and safety. It will naturally help promote cross-disciplinary education and well-rounded training of future IT workforce.The research problems tacked in this project are laid out as follows. First, built on promising preliminary results, this project will further push the limit of algorithmic performance considering various forms of data heterogeneity and communication constraints. The study will be extended to personalized and multi-model settings, transforming data heterogeneity from a challenge to an opportunity. The privacy of federated learning systems will be further enhanced through the rigorous framework of differential privacy, and multiple privacy protection tools will be effectively integrated to provide privacy in depth. During this process, learning performance, communication efficiency, and system robustness will be jointly considered, and their intricate relation with privacy will be examined. This project will culminate when relevant research problems are studied under a generalized federated learning framework, which further accommodates peer-to-peer distributed learning and assumes a multi-layer hierarchical structure. This new learning model can distribute machine learning across the device-to-edge-to-cloud continuum with flexible adaptation in learning structure, greatly expanding the application scope of federated learning in emerging wireless networks.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.