CCSS: Distributed Swarm Learning for Internet of Things at the Edge

点击翻译按钮获取中文摘要

With the vigorous growth of versatile Internet of Things (IoT) services, smart IoT devices are increasingly deployed at the edge of wireless IoT networks to perform collaborative machine learning tasks using locally collected data, giving rise to the edge learning paradigm. Because IoT networks have massive low-cost edge devices with limited capabilities and resources, IoT-driven edge learning faces major technical challenges caused by the communication bottleneck, data and device heterogeneity, non-convex optimization, privacy and security concerns, and dynamic operating environments. To overcome these challenges, this project builds a new framework of distributed swarm learning (DSL) through a holistic integration of artificial intelligence and biological swarm intelligence. The proposed DSL framework for edge learning is expected to benefit a wide range of IoT applications such as autonomous fleet management, massive wearable electronics, smart agriculture, to name a few. This project also provides broader societal impacts through student training, workforce development, research dissemination and outreach to minorities and local communities.This objective of this project is to develop an efficient distributed learning framework that coherently addresses the unique technical challenges related to swarm IoT with device restrictions and resource constraints on communication, computation and data, and in complicated edge environments with potential link failure, attacks, and topology changes. First, a new DSL framework is established by bridging federated learning with swarm optimization techniques. With theoretical backing, efficient information extraction and exchanging mechanisms are developed along with parsimonious transmission schemes for high efficiency in both communication and computation of model updates. Second, to cope with data heterogeneity, link failure and malicious attacks in practical IoT systems, robust DSL techniques are developed based on generative adversarial networks, multi-worker selection and analog transmission-and-aggregation techniques. Finally, to handle streaming data in an online fashion at the network edge, dynamic optimization techniques are investigated via the design of adaptive weights and exploration-exploitation strategies under the DSL framework. The outcomes of this research are expected to contribute to novel tools for learning and optimization tailored to real-time operation of large-scale IoT in dynamic environments, with technological impacts on statistical learning, signal processing and wireless communications.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.