I'm Aliasghar Khani

Machine Learning Researcher

Aliasghar Khani

Machine Learning Researcher

At the present time, I am enrolled in a master's degree program at Simon Fraser University under the supervision of Professor Ghassan Hamarneh. Also Dr. Saeid Asgari is my co-supervisor. I studied Computer Engineering at Sharif University of Technology (SUT) from 2016 to 2021. I conducted my B.S. project under the supervision of Professor MohammadHossein Rohban. This research project employed self-supervised and weakly supervised methods to improve adversarially trained models. My major interest and research area is computer vision and its fundamental problems like robustness and bias. In addition to it, I have hands on experience with language models and their combinition with computer vision. I love learning and using my knowledge for solving major real world problems both with a team and individually. Finally my research interests includes:

  • Robustness
  • Self supervised learning
  • Generalization

My Skills

Coding skills


Other Skills

Mathematics Knowledge
Machine Learning Knowledge
Machine Learning Frameworks


  • 2021 - Current
    Computer Science
    Simon Fraser University

    M.Sc. student under supervision of Professor Ghassan Hamarneh

  • 2016 - 2021
    Computer Engineering
    Sharif University of Technology

    B.Sc. in Computer Engineering

  • 2012 - 2016
    Mathematics and Physics
    Shahid Madani High School


  • 2022
    AI Research Internship

    Under the supervision of Dr. Saeid Asgari

  • 2020 - 2021
    Research at Bioinformatics and Computational Biology (BCB) Lab
    Sharif University of Technology

    Under the supervision of Prof. Mohammad Hossein Rohban

  • 2019 - 2021
    AI Researcher
    Iran's National Elites Foundation

    Under the supervision of Prof. Mahdieh Soleymani Baghshah


MaskTune : Mitigating Spurious Correlations by Forcing to Explore

This work proposes MaskTune, a masking strategy that prevents over-reliance on spurious (or a limited number of) features. MaskTune forces the trained model to explore new features during a single epoch finetuning by masking previously discovered features. MaskTune, unlike earlier approaches for mitigating shortcut learning, does not require any supervision.

Link to paper

Counterbalancing Teacher: Regularizing Batch Normalized Models for Robustness

In this work, we investigate low generalization performance for BN by first showing that removing BN layers across a wide range of architectures leads to lower out-of-domain and corruption errors at the cost of higher in-domain errors. We then propose Counterbalancing Teacher (CT), for increasing robustness.

Link to paper

MoNuSAC2020: A multi-organ nuclei segmentation and classification challenge

Automating the tasks of detecting, segmenting, and classifying nuclei can free up the pathologists' time for higher value tasks and reduce errors due to fatigue and subjectivity. We prepared a large and diverse dataset of nucleus boundary annotations and class labels and organized a challenge around this dataset. In this paper, we summarize the dataset and the key findings of the challenge.

Link to paper

A Deep Learning Framework for Viable Tumor Burden Estimation

In this paper, we propose a deep-learning framework for the segmentation of whole and viable tumor areas of liver cancer from whole-slide images (WSIs). To this end, we use a Fast Segmentation Convolutional Neural Network (Fast-SCNN) as our network. After data-augmentation on the training subset, we train the network with a multi-term loss function and SWA technique.

Link to paper

Towards automatic prostate gleason grading via deep convolutional neural networks

In terms of accuracy, microscopic inspection of biopsy tissues is best. The Gleason grading system is used to evaluate the stage of prostate cancer using prostate biopsy samples. It is time-consuming to grade each region in a tissue. In this paper, we propose an automatic method for this task based on a deep learning-based approach. The MobileNetV2 backbone is used with the DeepLabV3+ model.

Link to paper

A Unified Framework for Multi-organ Nuclei Segmentation and Classification Using Deep Convolutional Neural Networks

To handle the color variance in the images which originate from four organs, we use Vahadane stain normalization method. Then, we split the dataset into training and validation sets. Afterward, we add zero padding to the normalized images and patch them into tiles of 512× 512. We apply various augmentation techniques to the training set to avoid overfitting and add more data.

Link to paper

See my work

Information Retrieval System

Indexes Wikipedia's Data with text classification, clustering system, and a crawler that crawls pages and indexes in ElasticSearch.

Hospital Patients Reception Simulation

Simulator for the Reception of patients in a hospital.

IoT Monitoring Device

IoT device fro measuring soil temperature and humidity in the greenhouse and sends them to a web server.

Corona Virus Segmentation and Classification

Researched segmentation and classification of coronavirus in CT-scan images.

Cancerous Area Detection Website

Website for analyzing histopathology tissue images and detects cancerous portions.

Bag of Classifiers

We implemented Perceptron, MIRA, and Kernelized classifiers in python on MNIST and fashion datasets.