CAREER: Memory-Efficient, Heterogeneity-Aware and Robust Architecture for Federated Intelligence on Edge Devices

点击翻译按钮获取中文摘要

The recent trend of migrating computation from the centralized cloud to distributed edge devices is reshaping the landscape of today’s Internet, especially under the unprecedented challenges of the COVID19 pandemic. With privacy being a critical concern in data aggregation, Federated Learning emerges as a promising solution to such privacy-utility challenge. It pushes the computation towards consumer’s edge devices, where the data is generated. By exchanging statistical information, the participants perform collaborative learning in a distributed fashion. Unfortunately, the original design still faces new system-architectural challenges from limited memory, software/hardware heterogeneity, security and statistical diversity from different edge devices. The overarching goal of this CAREER project is to design, optimize and implement a memory-efficient, heterogeneity-aware and robust architecture for federated learning on consumer’s edge devices. In particular, it aims to: 1) remove the memory barriers of running the computational-intensive learning tasks; 2) resolve the software and hardware heterogeneity among various kinds of devices; 3) secure the information exchange and the machine learning backend. The research will provide a stack of solutions to address the urging needs in realizing collaborative intelligence on edge devices with computation/memory/energy-efficiencies, security and robustness. This research will address an urgent problem to bridge the gap between the vast data available from consumer’s edge devices and the rising interest of utilizing such private data to improve our wellbeing. The algorithms and tools developed in this CAREER project will lay the foundations to a plethora of new applications on massively distributed edge devices, as the essential elements for building a smart, connected and resilient community. The CAREER program will advance STEM education by developing new educational components related to machine learning, edge computing and security. This includes diverse outreach plans of cybersecurity summer camps, junior research symposium, high school instructor mentorship, coding competitions and the inclusion of underrepresented minority and women engineers. The potential use cases will be also explored with the collaborating industrial partners to enrich their business models.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.