Dr Maryam Sultana
Lecturer in Artificial Intelligence
School of Engineering, Computing and Mathematics
Role
Dr Maryam Sultana is a Lecturer in Artificial Intelligence in the School of Engineering, Computing and Mathematics at Oxford Brookes University. Her work sits at the intersection of computer vision and uncertainty quantification, with a particular focus on making deep learning systems aware of what they do not know. Alongside her lectureship she leads the research on developing uncertainty-aware perception for autonomous driving and engaging with potential industry partners.
She joined Oxford Brookes in 2022 as a Research Fellow on the EU Horizon 2020 Epistemic AI project, where she worked on epistemic uncertainty modelling, robust machine learning and trustworthy computer vision, and taught across the School's AI, machine learning and computer vision modules. Before that she was a Research Associate at the world’s first graduate level AI university, Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in United Arab Emirates, working on domain generalisation and moving object segmentation. She holds a PhD in Computer Science from KNU, South Korea (2021), with a thesis on robust foreground object segmentation using regularised generative adversarial networks, together with an MPhil and MSc in Electronics from Quaid-i-Azam University Pakistan.
Teaching and supervision
Modules taught
Dr Sultana teaches machine learning, machine vision and artificial intelligence at undergraduate and postgraduate level. Her recent teaching includes COMP7034 Machine Learning and Data Mining (PG), COMP6034 Advanced Machine Learning and Deep Learning, COMP6032 Artificial Intelligence (UG), COMP6011 Machine Learning (UG) and COMP6025 Machine Vision (UG). She supervises undergraduate and MSc dissertation projects in machine learning, computer vision and trustworthy AI. She is also part of two research groups VAIL and ADIT. She welcomes enquiries from prospective PhD students interested in uncertainty quantification, evidential and Bayesian deep learning, and robust visual perception.
Research
Her research develops principled ways of representing and reasoning about uncertainty in neural networks, drawing on belief functions, random sets, credal sets and Bayesian methods. Current projects include epistemic object detection, where detectors are trained with evidential heads that yield calibrated, decomposable measures of localisation and semantic uncertainty; post-hoc "epistemic wrapping" of Bayesian neural network posteriors into belief-function posteriors for second-order uncertainty; uncertainty-aware perception for autonomous vehicles within the U-Drive project. Earlier work established her contributions to domain generalisation with vision transformers and to adversarial and generative approaches to background subtraction and moving object segmentation and led to a US patent on self-distilled vision transformers for domain generalisation and a South Korean patent on GAN-based moving object detection.
She is a co-author of Credal Learning Theory (NeurIPS 2024) and of the book Epistemic Uncertainty in Artificial Intelligence, produced under the Epistemic AI project. Her work has appeared in Pattern Recognition, Neurocomputing, IEEE Transactions on Multimedia, Neural Networks, Machine Vision and Applications and ACCV, and has attracted over 1,000 citations. She collaborates with researchers at Oxford University in UK, TU Delft in Netherlands, KU Leuven in Belgium and through the Epistemic AI network.
Research interests:
-
epistemic uncertainty
-
trustworthy AI,
-
belief functions,
-
random sets
-
credal learning,
-
Bayesian deep learning,
-
Uncertainty aware object detection and segmentation
-
domain generalisation
-
vision transformers,
-
uncertainty-aware autonomous systems,
-
neural operators and scientific machine learning.