| Home | Research | Teaching | People | CV |
We study the framework of distributed machine learning within the multi-level hierarchy of user devices and edge servers, with an aim to develop new theories and techniques for communication efficiency, task scheduling, and learning system optimization. Examples research outcomes include efficient algorithms for ML task distribution among user devices, jointly optimal ML task distribution and communication resource allocation, and online scheduling and resource optimization methods.
This research targets the large-scale heterogeneous communication and networking architecture that serves as the backbone for the emerging mobile cloud/edge computing paradigm. We envisage a seamless global system of computing, communication, and applications, supported by a synergistically operated mobile cloud/edge computing system, incorporating macro cloud centers, micro cloudlets, edge computing hosts, and smart mobile devices. Example topics of investigation include mobile computation offloading, virtual machine placement and allocation, and co-operative joint communication-computation.
The goal of this research is to develop new theories, communication algorithms, and networking protocols for efficient allocation of spectrum, hardware, and power in high-throughput wireless networking environments. Topics of our investigation include co-operative communication, small-cell networks, interference management, stochastic optimization, and dynamic resource allocation.
The principles of network science permeate wide-ranging applications such as communications, cloud computing, power grid management, transportation, and biology. A central issue is how to effectively share network resources among competing agents. We are interested in developing new theories and practices for fair resource scheduling in large-scale networked systems. Examples of our investigation include cloud computing economics, distributed smart-grid control, and multi-resource fair scheduling.
We design adaptive machine learning techniques to categorize the network traffic, for quality-of-service provisioning and for anomaly detection. We take advantage of the emerging capabilities of computing at the network edge, where the network traffic is more localized with shared commonalities among local users, to improve the accuracy of traffic classification. We develop practical guidelines on how to design and operate network traffic classifiers, to optimally balance the trade-off between accuracy, privacy, cost, and delay.
(Links to electronic copies of all papers)
(Copyright notices)
Latest:
Google Scholar: https://scholar.google.com/citations?user=tSB5nzIAAAAJ