Mohammadreza Mohseni

I am a computer science MSc student at Simon Fraser University (SFU) supervised by Dr. Stella Atkins.

I'm passionate about theory, algorithms, and machine learning. At SFU and MetaOptima, I have worked on out-of-distribution detection to enhance AI-based decision making in dermatology. Also, I'm coaching SFU competitive programming teams in ACM-ICPC competitions. I did my Bacholor's at Shahid Beheshti University on computer engineering. I have been participating in the ACM-ICPC competitions during my undergrads and made it to the ACM-ICPC World-Finals twice.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github  /  Linkedin

• Our new preprint on "Out-of-Distribution Detection for Dermoscopic Image Classification" is available on arXiv.
• Our work on "Can Self-Training Identify Suspicious Ugly Duckling Lesions?" is accepted at ISIC Workshop at CVPR 2021.
• I have started to work part-time as a researh assistant in MetaOptima Technology Inc.

Do you think current out-of-distribution detection systems take false positives from all in-distribution classes equally? We show that the answer is NO.
Our BinanryHeads neural network fairly classifies dermoscopic skin images and detects novel disease images. We also propose a more realistic evaluation scheme for OOD detection systems.

Did you know skin lesions which look different than the others can be dangerous? These odd-looking lesions are called ugly ducklings.
In this work we show that a pipeline of detection, segmentation, and self-supervised outlier detection is able to detect suspicious ugly duckling lesions with the accuracy of over 94%.