CV

Education, research experience, publications, teaching, hardware and software skills. A PDF version is available via the icon on the right.

Contact Information

Name Seyed Mohammad Ojagh Kazazi
Professional Title M.Sc. Student in Electrical Engineering at EPFL
Email smohammadok@gmail.com
Location Lausanne, Switzerland

Professional Summary

Electrical engineer and M.Sc. student at EPFL working on computer vision, vision-language models, and efficient on-device AI. My work spans research, teaching, and hardware-aware systems, with a focus on models that remain accurate under real memory, latency, and reliability constraints.

Experience

  • 2026 - present

    Espoo, Finland (remote)

    Research Assistant
    Aalto University
    Supervisor: Prof. Siavash Khajavi.
    • Developed a knowledge-distillation framework combining KL-divergence and cross-entropy supervision to compress a 7B Qwen2.5 vision-language teacher into 0.5B–3B student models.
    • On DetectiumFire, the 0.5B student retained 95.2% of the teacher’s burning-object accuracy with 14× fewer parameters and reached 99.5% fire/non-fire accuracy after deployment.
    • Deployed and profiled distilled models on a Raspberry Pi 5 (4× Cortex-A76, 8 GB, CPU only): 1.9 GB peak memory, about 4 s end-to-end latency, and 2 false alarms on 1,452 test images.
    • First-author submission to AAAI 2027 Innovative Applications of AI (IAAI).
    • Currently investigating prompt optimization and reasoning strategies for Vision-Language-Action models.
  • 2024 - 2025

    British Columbia, Canada (remote)

    Research Assistant
    Raderon AI Laboratory
    Supervisor: Dr. S. Sarabi.
    • Conducted medical image analysis research combining the Discrete Wavelet Transform with swarm-based optimizers.
    • Co-authored a paper published in Elsevier Machine Learning with Applications.
  • 2024 - 2024

    Tehran, Iran

    Research Assistant
    Artificial Creature Laboratory (ACLab), Sharif University of Technology
    Supervisor: Prof. Saeed Bagheri Shouraki.
    • Designed and built a smart IoT socket for the lab’s custom AI platform, with a path toward commercialization.
    • Evaluated IoT platforms and network protocols, then implemented firmware on Arduino and ESP32/ESP8266.

Education

  • 2026 - present

    Lausanne, Switzerland

    M.Sc.
    École Polytechnique Fédérale de Lausanne (EPFL)
    Electrical Engineering
    • Fall 2026 coursework: Machine Learning, Mathematics of Data, Convex Optimization, Fundamentals of Inference and Learning.
  • 2021 - 2025

    Tehran, Iran

    B.Sc.
    Sharif University of Technology
    Electrical Engineering (Digital Systems)
    • GPA: 18.53/20 (3.95/4).
    • Top 15% of the B.Sc. Electrical Engineering class and top 3 in Digital Systems.
    • Selected graduate-level courses: Convex Optimization II (19.7), VLSI (19.1), Parallel Computing (18.4), Security in IoT (17.8), Deep Learning (17.2), Deep Generative Models (17.2).
    • Selected undergraduate courses: Linear Algebra / Mathematical Methods (20), Data Structures and Algorithms (19.5), Communication Systems (19), Introduction to Machine Learning (17.9), ASIC/FPGA System Design (17.2).

Teaching Experience

  • 2023 - 2024

    Tehran, Iran

    Laboratory Teaching Assistant
    Sharif University of Technology
    Logic Circuits and Digital Systems Lab, Fall 2023.
  • 2022 - 2025

    Tehran, Iran

    Teaching Assistant
    Sharif University of Technology
    Supported lectures, assignments, projects, and grading across electrical engineering and programming courses.
    • Engineering Probability and Statistics — Fall 2025
    • Logic Circuits and Digital Systems — four terms, 2023–2024
    • Mathematical Methods in Engineering — Spring 2024
    • Communication Systems — Spring 2024
    • Object-Oriented Programming — Spring 2023
    • Basic Programming — Fall 2022

Awards

  • 2021
    National University Entrance Exam (Konkour)
    National Entrance Examination

    Ranked in the top 0.1% of about 180,000 participants and admitted to Sharif University of Technology with a fully funded scholarship.

  • 2025
    Academic Standing
    Sharif University of Technology

    Top 15% cumulative GPA among B.Sc. Electrical Engineering students of the 2021 class, including top 3 in Digital Systems.

Publications

Skills

Machine Learning and Computer Vision: PyTorch, Hugging Face Transformers, OpenCV, Ultralytics YOLO, scikit-learn, VLMs/VLAs, generative models, knowledge distillation, object detection, multi-object tracking, on-device inference
Programming: Python, C/C++, CUDA, Verilog, Java, MATLAB, Assembly
Hardware and Systems: Vivado, Xilinx ISE, ModelSim, LTspice, Zynq FPGA, Raspberry Pi, Arduino, ESP32/ESP8266
Tools: NumPy, pandas, Git, Linux, MySQL, LaTeX

Languages

Persian: Native
English: Fluent (TOEFL iBT 105)
Spanish: A1

Interests

Research: Computer vision, vision-language and vision-language-action models, efficient and on-device deep learning, AI/FPGA accelerators for neural networks, GPU computing and parallel architectures, generative models, trustworthy machine learning
Personal: Football, volleyball, gym, travel, sitcoms, classical Persian music, guitar

Projects

  • Football Object Detection and Multi-Object Tracking

    Fine-tuned YOLOv8 on SportsMOT football video and built tracking with ByteTrack, DeepSORT, and CSRT, including occlusion and ID-switch analysis.

    • mAP@50 0.947, mAP@50–95 0.779, precision 0.93, recall 0.94; 1.4 ms inference per frame on a P100 GPU.
  • Membership Inference Attacks Against Diffusion Models

    Re-implemented SecMI (ICML 2023) on CIFAR-100 DDPMs and proposed a hyperparameter-free multi-timestep fusion variant.

    • Raised attack AUC from 0.971 to 0.984 and TPR@1% FPR from 0.519 to 0.642.
  • Machine Unlearning and Private Training

    Implemented SISA sharded training with ResNet-18 on CIFAR-10 and verified unlearning with backdoor success rate and membership inference.

  • Hardware and Parallel Computing

    CORDIC and FFT on Zynq FPGA (Verilog, Vivado), a multi-cycle MIPS processor, parallel merge sort and heat-transfer simulation with texture memory in CUDA, and a digital telecommunication system simulation.

References

  • Prof. R. Amiri 

        Associate Professor, Electrical Engineering Department, Sharif University of Technology — amiri@sharif.edu
  • Prof. S. Bagheri Shouraki 

        Full Professor, Electrical Engineering Department, Sharif University of Technology — bagheri-s@sharif.edu
  • Prof. S. Khajavi Haghighat 

        Assistant Professor, Industrial Engineering and Management, Aalto University — siavash.khajavi@aalto.fi
  • Dr. S. Sarabi 

        Director, Raderon AI Lab — s.sarabi@raderonlab.ca