Research

Undergraduate thesis project under the supervision of Dr. Rezwana Reaz. In this work, I co-developed an algorithm for dynamic task scheduling in fog systems, addresses the challenges of scheduling tasks with varying priority levels, deadline constraints, and dependencies utilizing the concepts of Directed Acyclic Graphs (DAGs), priority queues, and dynamic queue switching. My contributions included algorithm development by identifying limitations and making strategic adjustments in each iteration to enhance performance. I implemented simulation programs in Java and recommended configurations for experiments. Task lengths in our generated dataset were derived from the GoCJ dataset, which provides real-world data on various job characteristics. The simulation results showed significant improvements in response time, makespan, throughput, and task completion rate, especially for workloads with high task inter-dependencies.

This work is currently under review for publication.

Worked under the supervision of Dr. Shaikh Anowarul Fattah and Tanvir Mahmud [https://scholar.google.com/citations?user=4aZPxRsAAAAJ&hl=en]. In this project, we developed an efficient deep learning architecture with sequential 1D, 2D, and 3D feature extractors to capture intra- and interchannel dependencies in ECG signals, improving cardiovascular disease detection accuracy. We evaluated the architecture on a large publicly available dataset containing ECG signals from over 10,000 patients, achieving an accuracy of 97.3% in classifying six heartbeat rhythms. Additionally, we introduced a novel knowledge distillation framework that transferred knowledge from a high-performing teacher model to a student model, reducing the number of trainable parameters by up to eight times. The student model maintained strong performance while significantly improving efficiency and reducing complexity.

Published in IEEE Journal of Biomedical and Health Informatics DOI

Worked under the guidance of Khondoker Nazmoon Nabi. In this project, we compared the forecasting capabilities of several deep learning models, including CNN, LSTM, GRU, and multivariate CNN, for predicting COVID-19 cases in three countries. While RNN models are generally recognized for their effectiveness with time series data, in our experiments CNNs outperformed them by effectively learning local data patterns, as indicated by performance metrics such as MSE, nRMSE, and MAPE.

Published in Results in Physics DOI