Seminar: Machine Learning Seminar

ECE Women Community

Statistical Inference and Learning over Wireless Channels: Distributed Algorithms and Performance Analysis

Date: June,29,2022 Start Time: 10:30 - 11:30 Add to:
Lecturer: Prof. Koby Cohen
Affiliations: Electrical and Computer Engineering
At Ben-Gurion University
Research Areas:

Statistical inference and learning tasks often involve optimizing a global objective function in an uncertain environment. Traditional centralized algorithms, such as centralized gradient descent, or maximum likelihood require direct access to each sample in the dataset (e.g., when all user data is uploaded and processed at a central unit/server). With the increasing demand of data-intensive applications, however, centralized approaches become inefficient in terms of communication resources, storage, privacy, and computational aspects. Distributed learning and inference methods have attracted much attention in recent years to address this issue by distributing the computation among the nodes (e.g, smart mobile devices, sensor nodes, etc.), and making efficient learning and inference operations over the wireless communication channels.

In this talk, I will present several results from the research we conducted recently on this timely topic, focusing mainly on two problem settings. The first setting deals with distributed learning for dynamic spectrum access, which attracted much attention by academia and industry for future wireless communication, and was a main goal of a recent DARPA challenge. In this setting, I will present our recently developed online learning and deep reinforcement learning algorithms, and regret analysis. The second setting focuses on federated learning, which is a promising distributed machine learning framework studied in recent years, and was also implemented by Google Gboard App. In this setting, I will present our recently developed online learning and inference algorithms over multiple access channels, and finite-sample error analysis. Specifically, energy scaling laws for signal transmissions will be presented that guarantee the best possible centralized convergence rate.

* Koby Cohen is a Senior Lecturer (tenured Assistant Professor) at the School of Electrical and Computer Engineering at Ben-Gurion University of the Negev, Israel.


All Seminars
Skip to content