![Ehsan Ur Rahman Mohammed](http://www.uwindsor.ca/dailynews/msw/sites/uwindsor.ca.dailynews/files/styles/full/public/900_ehsan.jpg?itok=_GvC5Oli)
Science alumnus Ehsan Ur Rahman Mohammed (MSc 2023) is one of this year’s Governor General’s Gold Medallists for academic excellence at the graduate level and will be recognized during Convocation ceremonies on Oct. 12.
“It is very humbling because I was not expecting it, I felt extremely happy especially as an international student, I felt that my efforts in face of the challenges didn’t go unnoticed,” says Mohammed.
“I feel grateful to the university for providing me with the plethora of opportunities, starting from the admission process all the way to my graduation — everyone was extremely supportive. I don’t think it would be just if I don’t give credit to a couple of people in my journey at the University of Windsor, and they are Prof. Imran Ahmad, Dr. Tim Brunet, and the entire team at CTL, EPICentre, and Canterbury College.”
Under the supervision of Imran Ahmad, acting director of the School of Computer Science, Mohammed completed a master’s in computer science. His research revolved around finding solutions to generating data using artificial intelligence (AI) models when little data is available for training.
“For my research I was generating images out of the blue — making images of people who do not exist,” he says.
“Using generative adversarial networks, I was trying to combine two different concepts that hadn’t been combined earlier in literature i.e. transfer learning and attention mechanism.”
Mohammed says the original motivation was not to generate facial images, but was to come up with a procedure that would use off-the-shelf artificial intelligence models to generate medical images to train medical diagnostic models.
“For data scarce domains such as medical images, where it is hard to collect data due to privacy concerns, what I wanted to do is get a hold of pre-trained models and use those models to generate data,” he says.
Mohammed adds that another issue with medical images is their large size.
“If the research is pursued further then pre-trained models will not only help deal with patient’s privacy and confidentiality, but more importantly anyone who is going to create a medical models will not have to worry about collecting a lot of data to make their models robust,” says Mohammed.
“Collecting a lot of medical data is both expensive and time-consuming and involves many hurdles and setbacks such as privacy concerns, and lack of quality annotators.”
Without much labelled data, it makes it hard to create a robust model with all forms of shapes and intricate detail.
“But generating face images is a benchmark if you’re able to train face images you should be able to train almost any other kinds of images and with other improvements in future research, generating medical images would also be possible with ease,” says Mohommed.
During his master’s, says Dr. Ahmad, Mohommed co-authored two international peer-reviewed conference papers and one journal paper. He was the recipient of a Co-op/Internship Rising Star Student Award and won the final pitch competition of the RBC Founders Program.
“He also served as conference chair of the UWill Discover Conference 2023, head of events at the UWindsor AI Club, graduate student (MSc) representative, School of Computer Science and volunteered for many other activities,” says Ahmad.
“Ehsan has been an exemplary student with excellence in academics, research, and student leadership.”
In September 2023, Mohammed began doctoral studies at Western University. Working with two supervisors he will delve into the domain of deep learning by creating robust AI models that take inspiration from the human brain.