AI/ML Safety, Bias and Security

The aim of this project is to employ rigorous software engineering principles for designing robust decision making systems (e.g., robust and secure artificial intelligent and machine-learning systems). To this end, we focus on various desirable properties of decision making systems, including but not limited to security (e.g., resilience against adversarial and backdoor attacks), robustness and fairness (i.e., removing social discrimination). In this topic, we have made some of the pioneering contributions, specifically, to discover inherent bias in AI/ML software. In terms of practical impact, our proposed techniques for functional and fairness testing methodologies have discovered hundreds and thousands of bugs in well-established AI/ML models developed or used by popular software industries including but not limited to Google, Microsoft, IBM, Amazon and Airbnb.

Representative Publications:

Towards Backdoor Attacks and Defense in Robust Machine Learning Models
Ezekiel Soremekun, Sakshi Udeshi, and Sudipta Chattopadhyay
Elsevier Journal of Computers and Security, 2023

AequeVox: Automated Fairness Testing of Speech Recognition Systems
Sai Sathiesh Rajan, Sakshi Udeshi, and Sudipta Chattopadhyay
25th International Conference on Fundamental Approaches to Software Engineering (FASE), 2022

Astraea: Grammar-based fairness testing
Ezekiel Soremekun, Sakshi Sunil Udeshi, and Sudipta Chattopadhyay
IEEE Transactions on Software Engineering (TSE), 2022

Automated directed fairness testing
Sakshi Udeshi, Pryanshu Arora, and Sudipta Chattopadhyay
Proceedings of the 33rd (ACM/IEEE) International Conference on Automated Software Engineering (ASE), 2018

Grammar Based Directed Testing of Machine Learning Systems
Sakshi Udeshi and Sudipta Chattopadhyay
IEEE Transactions on Software Engineering (TSE), 2020

Interdisciplinary Publications:

Accented DH: Assessing Fairness of Multilingual Speech Recognition Systems
Setsuko Yokoyama, Sai Sathiesh Rajan and Sudipta Chattopadhyay
Digital Humanities (DH), 2023

Acknowledgement: We are grateful to Ministry of Education, Singapore, OneConnect Financial and Temasek Labs for generously supporting this project.


Ezekiel Olamide Soremekun
Assistant Professor (Royal Holloway, University of London)
Sai Sathiesh Rajan
PhD Student (SUTD), MS (Georgia Tech)