RII Track-4:NSF: HEAL: Heterogeneity-aware Efficient and Adaptive Learning at Clusters and Edges

With the proliferation of the Internet of Things (IoT) and technological advances in deep learning, various application domains have witnessed the growing adoption of Artificial Intelligence (AI), such as augmented reality, autonomous driving, and smart healthcare. Effectively learning from the ever-expanding pool of data generated by IoT devices poses a unique challenge due to data regulations and privacy concerns. Federated learning shows promise as a method for collaboratively training models on edge devices without exposing sensitive data. However, deploying federated learning in real-world IoT networks remains challenging due to the heterogeneity of systems and data, as well as the coexistence of multiple jobs. This project aims to address such challenges with a systematic solution, HEAL, for Heterogeneity-aware Efficient and Adaptive Learning for multiple jobs in a shared IoT network. The fellowship will provide support for the PI and her graduate student to conduct essential experimental investigations at the Coordinated Science Laboratory at the University of Illinois Urbana-Champaign, leveraging advanced cyberinfrastructure, cutting-edge technologies, diverse datasets, and abundant domain expertise at this interdisciplinary research institute. The project outcome will advance the knowledge and understanding of collaborative learning in the edge-cloud continuum and provide guidance for AI-driven applications on shared IoT infrastructure. This Research Infrastructure Improvement Track-4 EPSCoR Research Fellows (RII Track-4) project would provide a fellowship to an Assistant professor and training for a graduate student at the University of Louisiana at Lafayette. This work would be conducted in collaboration with researchers at the University of Illinois Urbana-Champaign. This project aims to address the unique challenges encountered when implementing practical federated learning in real-world IoT networks, catering to the growing diversity of machine learning applications. A systematic solution will be designed for Heterogeneity-aware Efficient and Adaptive Learning (HEAL) for multiple jobs in a shared IoT network, synergizing two major thrusts: adaptive offloading of on-device training computation to the edge server and judicious selection and scheduling of participant devices for concurrent learning jobs. It will systematically and experimentally investigate a number of knotty issues in multi-job federated learning in a shared heterogeneous IoT infrastructure. The major components with novelty are: (1) The adaptive offloading of training computation from heterogeneous edge devices that can strike a balance between computation, communication, and privacy leakage risk; and (2) The judicious coordination of edge devices in the distributed training procedures of multiple concurrent learning jobs, aiming for system efficiency and model quality. The proposed solution will be deployed and tested in real-world IoT networks at the host site over diverse learning applications, not only providing solutions at the algorithmic level but also producing practical implications and insights. The anticipated project outcomes will enrich educational materials and strengthen curriculum development in the areas of machine learning systems, distributed systems and networking, cloud and edge computing, and resource scheduling. This project will enable the PI to establish a long-term collaboration with national prominence and enhance the research capacity of her home institution. 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.