Mohamadreza Khanmohamadi

Computer Vision & AI Researcher

I am a deep learning and computer vision researcher, currently completing my B.Sc. in Computer Science at Shahid Beheshti University. My research focuses on interactive segmentation, distributed inference for Mixture-of-Experts models, and Reinforcement Learning from Human Feedback (RLHF). I have collaborated with academic institutions such as UCLA and the University of Tehran, and contributed to several publications in top venues like IEEE COINS and Springer Nature journals.

I am particularly passionate about deploying scalable AI systems for real-time applications, and my work spans probabilistic modeling, symbolic mathematics using GNNs, and active learning. I also contribute as a teaching assistant in Data Science and Deep Learning courses.

Email Google Scholar GitHub Twitter LinkedIn
Mohamadreza Khanmohamadi

About Me

Education

Shahid Beheshti University

Department of Computer Science
B.Sc. in Computer Science
GPA: 3.76/4 (Top 3 of 55 students)
Thesis: Design and Implementation of RLHF System for Interactive Segmentation

Sept. 2020 – Dec. 2024

Awards & Achievements

Best Paper Nomination – IEEE COINS

2024

1st Place, Innotech AI Competition (Fraud Detection)

2023

Ranked Top 1% in National University Entrance Exam (Iran)

2020

Research Experience

Remote Research Collaboration

UCLA

Supervisors: Dr. Nader Sehatbakhsh, Dr. Farshad Firouzi

Working on distributed inference of Mixture of Experts (MoE) for edge computing and adaptive real-time video segmentation.

Nov 2024 – Present

Computer Vision Researcher

University of Tehran - Applied Artificial Intelligence Lab (AAIL)

Supervisor: Prof. Mohammad Ali Akhaee

Developed interactive segmentation models with RLHF using Deep Deterministic Policy Gradient (DDPG) on BRATS dataset.

Nov 2022 – Jan 2025

Research Internship

Computational Mathematics and AI Lab, SBU

Supervisor: Prof. Kourosh Parand

Designed Graph Neural Networks to solve symbolic mathematics problems, achieving 81% accuracy.

May 2022 – Sep 2022

Publications

InSureNet: A Graph Transformer Network Approach for Health Insurance Fraud DetectionSpotlight

IEEE COINS 2025 Conference 2025

Healthcare insurance fraud imposes a significant financial and operational burden on medical systems, necessitating robust and scalable detection methods. This paper presents InSureNet, a novel framework based on Graph Transformer Networks (GTNs) for provider-level fraud detection. Unlike conventional graph neural networks with fixed connectivity, GTNs automatically discover and learn compositional meta- paths, enabling the model to capture multi-hop, heterogeneous relationships embedded in healthcare data. By integrating meta-path induction with graph convolution, InSureNet effectively models higher-order dependencies and structural patterns among providers. We evaluate our approach on a publicly available healthcare fraud dataset, demonstrating that it outperforms state-of-the-arts. Specifically, InSureNet achieves an improvement of 6% in F1-score and 17% in ROC-AUC, highlighting its ability to detect complex fraud scenarios under class imbalance.

Aref Dorahaki, Mohamadreza Khanmohamadi, Bahar Farahani

DETR-SAM: Automated Few-Shot Segmentation with Detection Transformer and Keypoint Matching

IEEE-COINS 2024

Few-shot Segmentation (FSS) aims to segment objects in an image using only a few annotated examples. The Segment Anything Model (SAM) has recently gained attention in FSS due to its versatility and capability to handle various segmentation tasks with prompts...

Mohamadreza Khanmohamadi*, Bahar Farahani*

Paper

Improving spleen segmentation in ultrasound images using a hybrid deep learning framework

Nature Publishing Group UK - Scientific Reports 2024

This paper introduces a novel method for spleen segmentation in ultrasound images, using a two-phase training approach. In the first phase, the SegFormerB0 network is trained to provide an initial segmentation...

Ali Karimi*, Javad Seraj*, Fatemeh Mirzadeh Sarcheshmeh, Kasra Fazli, Amirali Seraj, Parisa Eslami, Mohamadreza Khanmohamadi*, et al.

Paper

Smooth Adaptive Saliency-Based Training Endoscopy Images ClassificationSpotlight

Under review - Discover Artificial Intelligence (Springer) 2024

In the realm of artificial intelligence (AI), a significant challenge is the prevailing lack of trust, particularly stemming from the opacity of how these models operate. This issue becomes especially critical in the medical domain...

Amirsadegh Roshanzamira, Mohamadreza Khanmohamadi, Kasra Fazli Zaman Abadib, Giti Javidic, Ehsan Sheybanic

Code

Design and implementation of a novel RLHF system for glioma interactive segmentationSpotlight

Bachelor Thesis 2024

The goal of brain tumor segmentation is to aid in surgical planning. Traditional medical approaches focus on cellular differences, while AI methods, especially deep learning, process medical images...

Mohamadreza Khanmohamadi

Code