Mehil B. Shah
PhD Student at Dalhousie University
Faculty of CS
Dalhousie University
Halifax, NS
I am a PhD Student in the Faculty of Computer Science at Dalhousie University. I am currently being advised by Prof. Masud Rahman, and Prof. Foutse Khomh. My primary research interests are at the intersection of Software Engineering and Deep Learning, with a focus on secure, reliable, and explainable deep learning systems. My current research focuses on understanding the behaviour of bugs in deep learning systems and trying to improve the reproducibility of deep learning bugs.
My research interests focus on developing effective techniques to debug data bugs throughout the entire lifecycle in deep learning systems used for software engineering tasks. Specifically, I am interested in investigating bugs from errors in collecting, labelling, and preprocessing code, text, metric, and graph-based training data. My goals are to leverage existing software engineering research, large language models, and explainable AI (XAI) techniques to explain and accurately localize data defects, develop techniques to reproduce bugs in sandbox environments, and propose methods to repair data through model-data interactions. Overall, my research aims to enhance the reliability, robustness, and trustworthiness of safety-critical deep learning systems by addressing the key challenges in effectively debugging data bugs end-to-end.
news
Oct, 2024 | Our journal article titled “Towards Enhancing the Reproducibility of Deep Learning Bugs: An Empirical Study” is accepted by EMSE. The preprint is available on arXiv. |
---|---|
Jul, 2024 | We received Major Revision for our collaborative work titled “Towards Understanding the Challenges of Bug Localization in Deep Learning Systems” from EMSE. |
Jun, 2024 | Will be serving as the Web Chair for the SANER 2025. Thanks to Prof. Foutse Khomh, Prof. Mohammad Hamdaqa and Prof. Masud Rahman for the opportunity! |