Hossein Jafarinia

Hossein Jafarinia

Grad, SUT

Welcome

My name is Hossein Jafarinia

I am a researcher with a deep interest in Computer Vision, Medical Image Analysis, and Digital Pathology. My work has primarily revolved around developing novel architectures and methods for Whole Slide Image (WSI) Classification and Multiple Instance Learning (MIL), with applications in cancer detection and localization. Over the course of my academic journey, I have collaborated closely with research teams, mentored interns, and contributed to significant publications in my field.

Education

Sharif University of Technology (SUT), Tehran, Iran

M.S. in Computer Engineering (2021 – 2024)

  • Thesis: Classification and Localization of Cancer in Histology Images Using Weak Label
    Developed a new framework featuring a novel sparse transformer-based MIL-pooling method, which achieved state-of-the-art results in WSI classification.

Bu-Ali Sina University (BASU), Hamedan, Iran

B.S. in Computer Engineering (2015 – 2020)

  • Final Project: Detection of Infected Blood Cells by Malaria
    Designed and implemented a model for detecting malaria-infected blood cells, contributing to automated diagnostics.

Publications

Published

  • “The Silent Helper: How Implicit Regularization Enhances Group Robustness.” Analyzing the implicit regularization in training in the presence of spurious correlation.

  • “Snuffy: Efficient Whole Slide Image Classifier”. Accepted for ECCV 2024.
    Achieved state-of-the-art results in WSI classification on CAMELYON16 and TCGA Lung Cancer datasets.

  • “MILFORMER: Weighted Dual Stream Class Centered Random Attention Multiple Instance Learning For Whole Slide Image Classification”. Accepted for AAAI W3PHIAI Workshop 2024.
    Introduced a novel architecture for WSI classification, achieving top performance in classification tasks.

Submitted

  • “Navigating the MIL Trade-Off: Flexible Pooling for Whole-Slide Image Classification.” Focusing on Whole Slide Image classification with an emphasis on representation and classical methods.

Under Preparation

  • “Low-resource Persian Languages for Intelligent Systems”. Under preparation.
    Developed linguistic materials for Isfahani, a low-resource Persian dialect, and managed resources for other underrepresented Persian languages as part of an effort to document and publish them.

Projects

  • Contrastive Language-Image Pretraining (CLIP) for Histology WSIs
    Adapted OpenAI’s CLIP model for histology whole slide image classification, enabling the model to leverage text-image pairings for improved image understanding.

  • Cancer Mutation Detection with Microarray Data
    Developed a machine learning pipeline for detecting cancer mutations based on microarray gene expression data.

  • Small Molecule Quantum Feature Extraction and Target Protein Docking
    Focused on quantum feature extraction in small molecules and performed docking simulations to predict interactions between target proteins and small molecules.