Alireza Dizaji

About Me

(Last Updated: 5th Oct 2023)

CV: PDF

I’m Alireza Dizaji and I’m now settled in Montreal, Quebec. I’m doing my first year Ph.D. at Quebec Artificial Intelligence Institute, shortly known as MILA, and at Universite de Montreal. Before here, I graduated as a Bachelor of Science student at the Sharif University of Technology, Computer Engineering field; I have 3+ years of research experience in machine learning; Between May 2022 and 2023, I was an intern at the Technical University of Munich working on two projects: 1) Gait-disorder recognition for video-series skeleton graphs using semi-supervised methods and 2) benchmarking explainability methods at Graph Neural Networks, under the supervision of Prof. Daniel Rueckert and Nassir Navab. Between April 2021 - December 2022, I was an R&D scientist at CommaMed, where my main task was to develop AI methods for breast cancer detection. My research interests are Computer vision, Graph Neural Networks, Tensor Networks, and Explainable Machine Learning.

Publications

How to Evaluate Your Graph Neural Network Explanation by Retraining


Currently, several metrics have been proposed to evaluate the faithfulness of current state-of-the-art explanation methods in graph neural networks. However, there are several drawbacks among these evaluation metrics that make their fair comparison challenging. Through this project, we mentioned these drawbacks and instead proposed novel solutions with a guideline assisting in analyzing the focus of these explainers better, and the target Graph Neural Networks.

Alireza Dizaji, Razieh Rezaei, Anees Kazi, Nassir Navab, Ashkan Khakzar, Daniel Rueckert

LAP:An Attention-Based Module for Faithful Interpretation and Knowledge Injection in Convolutional Neural Networks


Through this project, we devised a self-interpretable model which could be easily injected into any CNN network. The method is a novel inspiration of both pooling and attention-based methods representing promising results on many experiments passing state-of-the-art methods.

Rassa Ghavami Modegh, Ahmad Salimi, Alireza Dizaji, Hamid R. Rabiee

An Artificial Intelligence-Based Computer Aided Detection Tool Helping Radiologists in Reading Digital Mammograms. (Medical publication)


Breast cancer is considered the deadliest one among women. Our main goal is to provide an AI assistant product to help radiologists with more accurate and swifter decision-making through mammography screening and therefore increase the survival rate of our Iranian mothers and ladies. As a member of the scientific team, I have contributed to a huge project containing multiple sub-branches, including pre and post-image processing, malignancy classification, and lesion segmentation, using mostly weakly and semi-supervised methods due to insufficient annotated data.

Nasrin Ahmadinejad (Radiologist), Nahid Sadighi (Radiologist), Rassa Ghavami Modegh, Alireza Dizaji, Amin Rezaei, Maryam Rahmani, Arvin Arian, Salome Maghsudlu, Saeedeh Shokri, Niloofar Pashaeifakhri, Mehran Arab Ahmadi, Ali Sefidmouy, Mahdi Ghaznavi, Amir Tofighi Zavareh, Hamed Dashti, Hamid R. Rabiee, Masoumeh Gity

Experience

Quebec Artificial Intelligence institute (MILA)

Graduate Research Assistant

Sep. 2023 - present

Technical University of Munich

Undergraduate Research Assistant

Dec 2022 - May 2023

I was working on developing semi-supervised methods to recognize the type of gait disorders of patients using video-series skeleton data. The main challenge here was the data itself. Due to the high number of unlabeled data and a considerable amount of missing points of multiple video-series samples, we tried to develop a semi-supervised method to leverage most of these samples.

Technical University of Munich

Research Intern

May 2022 - Nov 2022

Through those seven months, I have been working under the co-supervision of Prof. Navab and Prof. Rueckert, where our main focus is to analyze the drawbacks of current evaluation metrics for explanation methods in graph neural networks, and instead devise a new evaluation metric to better represent the faithfulness of these explainers.

CommaMed (formerly known as AI-Med)

R&D Scientist

April 2021 - December 2022

For more than a year and a half, I have been collaborating with the CommaMed start-up, which our goal is to both accelerate and increase the accuracy of diagnosis by devising novel AI methods. During this duration:

  • We have been developing multiple semi/weak supervision methods to fill the absence of sufficient data, and our framework is capable of diagnosing malignancy and localizing highly risk abnormalities with low miss rates.
  • In addition, I have contributed to the mentorship and management of interns, guiding them through a sub-branch of projects including devising and implementing more semi-supervised methods with higher accuracy.
  • For some reasons, including making run-time faster, I have also transported our python codes to C++, using Libtorch.

Sharif University of Technology

Research Assistant

August 2020 - March 2021

In August, I joined DML lab as a Research assistant where I was working on the impact of contrastive learning approaches on deep reinforcement-learning algorithms. to be more specific, I implemented CURL and SIMCLR (via Pytorch) algorithms to improve the movement of an agent within the Maze environment.

Sharif University of Technology

Teaching Assistant

Februrary 2019 - July 2021

Since my second semester, I have collaborated as a teaching assistant with several faculty members of computer engineering department. All courses that I have contributed to are listed below:

Spring 2021

  • Machine learning (M.Sc. course), Dr. Rohban
  • Modern information retrival, Dr. Soleymani

Fall 2020

  • Artificial intelligence, Dr. Rohban
  • Linear algebra, Prof. Rabiee

spring 2020

  • Computer architecture, Prof. Asadi
  • Data structures and algorithms, Dr. Safarnejad broujeni

spring 2019

  • Advanced programming, Mr. Hatami

Education

MILA / Université de Montréal

Ph.D. Computer Science (DIRO)

Sep. 2023 - present

Sharif University of Technology

B.Sc. Computer Engineering

Sep 2017 - Sep 2022

Studying Computer Engineering at Computer Engineering Department. GPA: 17.93/20 (last six semesters: 18.51/20)

Related courses

Convolutional neural networks for visual recognition (audited, online Stanford), machine learning (audited, online Stanford), artificial intelligence, linear algebra, data structures and algorithms, introduction to bioinformatics, modern information retrieval, design of algorithms, probability and statistics, discrete structures.

Allameh Amini highschool

Diploma

2013-2017

I received my Diploma in Mathematics and Physics from Allame Amini highschool, where I was ranked 24th out of 350k students in Iranian nationwide mathematics exam (Konkur).