Embedded Files
"Efficient Regression Models for Scan Statistics" arXiv 2026
This project introduces a class of scan statistics that detect interval anomalies in signals with smoothly varying backgrounds. Classical scan statistics assume a constant background, which fails on scientific signals that drift by design. We extend the framework to polynomial and Nadaraya-Watson kernel regression backgrounds, and develop incremental-update algorithms that reduce the cost from quartic to linear in signal length. Applied to ALMA radio telescope bandpass calibration, the method identifies every confirmed platforming defect in a set of 38,881 spectra, against roughly 50% missed by existing manual and automated review.
This project aims to assess public perception of the prevalence of filtered images on social media and to examine the use of automated assistance to recognize them and determine the extent of filtering applied. We developed InnerEye, an automated tool that provides both qualitative and quantitative analyses of the extent of filtering applied to an image. Additionally, we conducted a user survey to evaluate the tool's effectiveness and usability.
In this project, we designed an Electronic Health Record (EHR) storage system that utilizes blockchain and off-chain technologies to store individuals' COVID-19 status, enabling verification before granting access to public places. Furthermore, we integrated a deep learning-based facial recognition system to ensure the integrity of health certificates.
"Security Code Review Analysis & Automated Detection" BUET Undergrad Thesis 2019
In this project, we conducted qualitative and quantitative analyses of the prevalence of security code review in a dataset of code reviews collected from the Chromium Gerrit and annotated by expert researchers. We also built a security code review detection pipeline based on classical machine learning algorithms and improved the results using the boosting framework LightGBM.
Page updated
Google Sites
Report abuse