September 16, 2022

About the speaker

Mr. S M Sarwar, who is a Graduate Student majoring in Computer Science from the College of Engineering and Computer Science, is a recipient of the Presidential Research Fellowship.

S M is now pursuing his master's degree in Computer Science and working as a Graduate Research Assistant at The University of Texas Rio Grande Valley (UTRGV). S M received his Bachelor of Science and Master of Science degrees in Computer Science and Engineering from Jahangirnagar University in Bangladesh, respectively October 2019 and March 2022.

After joining UTRGV in Fall 2021, he was elected as the Secretary of Engineers Without Borders at UTRGV. Moreover, he has a membership in 10 other organizations within UTRGV. In spring 2022, He was elected as Vice President for Collegiate Entrepreneurs’ Organization – UTRGV, and in summer 2022, he became its President. In a short time, S M has made a tremendous impact in leading the organization and preparing CEO - UTRGV chapter for the future. In addition, S M was elected as a Graduate Senator for UTRGV's Student Government Association (SGA) for the 2022 – 2023 academic year. He is a member of the academic affair committee of SGA and closely working to establish a Graduate Students Council (GSC). Till now, he has participated in ICML'22, HSI Battle of the brain 22, UTRGV NSF I-Corps program 22, UTRGV – Engaged Scholar Symposium '22, and some other research-related conferences.

Outside school, he volunteers to organize events and track metrics as a Mozilla Representative for two Mozilla communities in India. Additionally, he has started serving as a member of the Youth Advisory Council of Congressman Mr. Gonzalez for the 2022-2023 academic year. Furthermore, starting summer of 2022, he started working as a Vice-Chair at IEEE Corpus Christi Section as community service.

Trade-off between federated learning and hierarchical federated learning

Federated learning is a collective learning process where we can train a model without exchanging users' original data. For the traditional cloud-based FL model, there exists only one central cloud-based server and N clients. The FAVG algorithm processes and aggregates after every step of gradient descent on each client to reduce the overall communication cost. The process is ongoing until the model achieves the desired accuracy or the limited resources, e.g., the communication or time budget, run out. On the other hand, for hierarchical federated learning, the HierFAVG algorithm is used to process the whole model. The main difference between FAVG and HierFAVG is edge-based server and cloud-based server systems. At hierarchical federated learning, we have to use both, but we can use only cloud servers for cloud-based. Another significant difference is that there is no connection between cloud servers and clients at hierarchical FL. Edge-server handles its clients. When clients send an updated model to edge-server, then it sends to the central server.