About
The growth of devices at the network edge — both in number and in computational ability — and the data generated by these devices has enabled the pervasive development of machine learning-powered applications, especially on those devices. While the cloud-based, centralized machine learning and artificial intelligence has been the dominant paradigm for these paradigms, recent trends towards better privacy, more efficient communication, and more robustness and resilience have encouraged and motivated the emergence of a new paradigm that pushes machine learning and artificial intelligence to the edge — edge intelligence. Edge intelligence is commonly described as the confluence of edge computing and artificial intelligence (Deng et al., 2019), and so it inherits many of the benefits and challenges of its derived fields.
In our research, we work in the space of fully decentralized in-edge learning (Zhou et al., 2019), looking specifically at how agents in a decentralized learning network can coalesce into dynamic learning communities, based on affinities in the network infrastructure or in the agents’ data or metadata. Our intuition — validated by our preliminary results (Kuttivelil and Obraczka, 2021) — is that by conducting model sharing and aggregation between these affinity-based learning communities, decentralized learning communities can achieve better, more differentiated model performances with significantly less communication overhead. In addition to using data and metadata, we are also developing application-agnostic methods for clustering based loss surface-based clustering and other works done on centralized federated learning. We are employing our proposed methods in a variety of real-world applications to confront challenges in managing heterogeneity, reliability, resilience, and robustness — including natural language processing, intelligent sensor networks for monitoring natural environments, and distributed social networks.
The development of our collaborative, community-structured decentralized edge intelligence systems necessitates the need for proper simulation platforms. This has resulted in parallel development of our own simulation tools, some built specifically for decentralized learning and others that are more general-purpose, such as our simple but powerfully extensible tool to bridge applications, such as our decentralized learning applications, to network simulators — the Network Simulation Bridge, or NSB.
Contact Us
If you’re interested in working with us on this research, please contact Ph.D. student Hari Kuttivelil (hkuttive@ucsc.edu) or Professor Katia Obraczka (katia@soe.ucsc.edu).
Resources on Edge Intelligence
We’ve also included some resources to get to know the emerging field of edge intelligence and various aspects of it, such as edge learning paradigms, applications, and more.
Academic Papers
- T. Rausch and S. Dustdar, “Edge Intelligence: The Convergence of Humans, Things, and AI,” in 2019 IEEE International Conference on Cloud Engineering (IC2E), Jun. 2019, pp. 86–96. doi: 10.1109/IC2E.2019.00022.
- M. Satyanarayanan, “The Emergence of Edge Computing,” Computer, vol. 50, no. 1, pp. 30–39, Jan. 2017, doi: 10.1109/MC.2017.9.
- S. Deng, H. Zhao, W. Fang, J. Yin, S. Dustdar, and A. Y. Zomaya, “Edge Intelligence: The Confluence of Edge Computing and Artificial Intelligence,” IEEE Internet of Things Journal, vol. 7, no. 8, pp. 7457–7469, Aug. 2020, doi: 10.1109/JIOT.2020.2984887.
- Z. Zhou, X. Chen, E. Li, L. Zeng, K. Luo, and J. Zhang, “Edge Intelligence: Paving the Last Mile of Artificial Intelligence With Edge Computing,” Proceedings of the IEEE, vol. 107, no. 8, pp. 1738–1762, Aug. 2019, doi: 10.1109/JPROC.2019.2918951.
- D. Xu et al., “Edge Intelligence: Architectures, Challenges, and Applications,” Mar. 2020, Accessed: Mar. 29, 2021. [Online]. Available: https://arxiv.org/abs/2003.12172v2
- Y. Shi, K. Yang, T. Jiang, J. Zhang, and K. B. Letaief, “Communication-Efficient Edge AI: Algorithms and Systems,” IEEE Communications Surveys Tutorials, vol. 22, no. 4, pp. 2167–2191, Fourthquarter 2020, doi: 10.1109/COMST.2020.3007787.
People
Harikrishna Kuttivelil, Ph.D. Student
Professor Katia Obraczka, Research Advisor