Sahar Dastani

I am a PhD Student at Mila/ÉTS, where I am working on Spatio-temporal video processing with applications to medical data. I did my Masters and Bachelors from Amirkabir University of Technology (AUT) where I was advised by Dr. M.Hassan Shirali-Shahreza and Dr. Mohammad Akbari. My Master thesis was about 3D Human Pose Estimation using Neural Networks. I also worked on developing Recommender Systems for Holoo accounting and financial software.

Email / CV / linkedin / Github

Lates News

Our abstract Psychomotor Agitation Detection using Deep Learning got accepted at WFPICCS.
I awarded the Upper Bound Talent Bursary 2024.
I awarded the Professor Cho Diversity scholarship.
I served as a Teaching Assistant for the Representation Learning course at Udem, under the direction of Aishwarya Agrawal.
I was invited by CIFAR to serve as a program reporter for the DLRL summer school in 2023.
Our paper RelationalUNet for Image Segmentation got accepted at MLMI.
I graduated from AUT with a Master's degree in computer science with GPA 4/4. (August 2022)
I joined Holoo Software Company as a Machine Learning Specialist from summer 2021.
I joined M.Hassan Shirali-Shahreza and Mohammad Akbari's lab as a Master student at AUT from Fall 2020.
I graduated from AUT with a bachelor's degree in computer science. (August 2020)

Publications

Psychomotor agitation detection using deep learning.
Sahar Dastani, Ali Harakeh, Arnaud Wiedemann, Phillipe Jouvet, Samira E. Kahou, WFPICCS' 24
We propose an automated video analysis system for a Pediatric Intensive Care Unit using deep learning to assess anxiety and distress indicators in pediatric patients during medical transportation.

Relational UNet for Image Segmentation
Ivaxi Sheth, Pedro Braga, Shivakanth Sujit, Sahar Dastani, Samira Ebrahimi Kahou
International Workshop on Machine Learning in Medical Imaging 2023
[ Paper , Code ]

We propose RelationalUNet which introduces relational feature transformation to the UNet architecture. RelationalUNet models the dynamics between visual and depth dimensions of a 3D medical image by introducing Relational Self-Attention blocks in skip connections.

Research

I'm interested in deep learning, self-supervised learning, out-of-distribution generalization.

Source code and style from Jon Barron's website.