Mohammad Kazzazi

EPFL

Lausanne, Switzerland

I am a master’s student in Electrical Engineering at EPFL, Lausanne. My research interests are in computer vision and multimodal learning, with a focus on making vision-language models efficient enough to run on real devices without losing their ability to reason about a scene.

Currently, I work with Prof. Siavash Khajavi at Aalto University on knowledge distillation of vision-language models for on-device fire understanding, and on prompt optimization and reasoning for vision-language-action models. Before that, I worked on medical image analysis at Raderon AI Laboratory with Dr. Soroush Sarabi, and on embedded IoT systems at the Artificial Creature Laboratory with Prof. Saeed Bagheri Shouraki.

I received my B.Sc. in Electrical Engineering (Digital Systems) from Sharif University of Technology, where I wrote my thesis on closed-form localization with Prof. Rouhollah Amiri and served as a teaching assistant for courses in programming, digital systems, and signals. My background in FPGA and GPU programming shapes how I think about deploying models.

Outside research, I enjoy football, volleyball, travel, and playing guitar.

Research interests

Computer vision and multimodal learning
Vision-language and vision-language-action models
Efficient and on-device deep learning
Generative models
Trustworthy machine learning
Hardware acceleration on GPUs and FPGAs

Research experience

Apr 2026 – present Research Assistant, Aalto University (remote) · advised by Prof. Siavash Khajavi
Distilling vision-language models for on-device fire understanding; reasoning in vision-language-action models.
Aug 2024 – Jan 2025 Research Assistant, Raderon AI Laboratory (remote) · advised by Dr. Soroush Sarabi
Medical image analysis with discrete wavelet transforms and swarm-based optimizers.
Jul – Sep 2024 Research Assistant, Artificial Creature Laboratory, Sharif University of Technology · advised by Prof. Saeed Bagheri Shouraki
Smart IoT socket on Arduino and ESP32/ESP8266 for the lab's AI platform.
2024 – 2025 B.Sc. thesis, Sharif University of Technology · advised by Prof. Rouhollah Amiri
Closed-form TDOA/FDOA estimation for moving-source IoT localization.

news

Sep 14, 2026	Started my M.Sc. in Electrical Engineering at EPFL in Lausanne.
Sep 05, 2026	Our paper, Distilling Vision-Language Models for On-Device Fire Understanding, is now on arXiv.
Mar 31, 2026	Accepted to the M.Sc. in Electrical and Electronic Engineering at EPFL.
Mar 01, 2026	Our paper, Enhancing skin cancer diagnosis using late discrete wavelet transform and new swarm-based optimizers, was accepted in Machine Learning with Applications.

selected publications

arXiv
Distilling Vision-Language Models for On-Device Fire Understanding

Mohammad Kazzazi, Zixuan Liu, and Siavash Khajavi

arXiv preprint arXiv:2609.05782, Sep 2026

Abs DOI arXiv Bib HTML PDF

Vision-language models (VLMs) offer a promising alternative to conventional fire detection systems by reasoning about the semantic context of a scene and thus reducing false alarms, yet their large model size makes deployment on embedded fire sensors impractical. We develop a teacher-student knowledge distillation framework in which large VLMs fine-tuned for fire understanding are distilled into lightweight students, then deploy the distilled models on a commercial Detectium fire detection sensor and evaluate accuracy, latency, and memory. Compact students preserve most of their teachers’ fire-understanding capability, and Qwen2.5-0.5B provides the strongest overall deployment trade-off.
@article{kazzazi2026distill, title = {Distilling Vision-Language Models for On-Device Fire Understanding}, author = {Kazzazi, Mohammad and Liu, Zixuan and Khajavi, Siavash}, journal = {arXiv preprint arXiv:2609.05782}, year = {2026}, month = sep, archiveprefix = {arXiv}, primaryclass = {cs.AI}, doi = {10.48550/arXiv.2609.05782}, url = {https://arxiv.org/abs/2609.05782}, }

MLWA

Enhancing skin cancer diagnosis using late discrete wavelet transform and new swarm-based optimizers

Ramin Mousa, Saeed Chamani, Mohammad Morsali, and 3 more authors

Machine Learning with Applications, Mar 2026

Abs DOI Bib HTML

@article{kazzazi2026skin,
  title = {Enhancing skin cancer diagnosis using late discrete wavelet transform and new swarm-based optimizers},
  author = {Mousa, Ramin and Chamani, Saeed and Morsali, Mohammad and Kazzazi, Mohammad and Hatami, Parsa and Sarabi, Soroush},
  journal = {Machine Learning with Applications},
  volume = {23},
  pages = {100811},
  year = {2026},
  month = mar,
  publisher = {Elsevier},
  doi = {10.1016/j.mlwa.2025.100811},
  url = {https://doi.org/10.1016/j.mlwa.2025.100811},
}

arXiv
CLEAR: A Closed-Form Minimal-Sensor TDOA/FDOA Estimator for Moving-Source IoT Localization

Mohammad Kazzazi, Mohammad Morsali, and Rouhollah Amiri

arXiv preprint arXiv:2510.04160, Oct 2025

Abs DOI arXiv Bib HTML PDF

CLEAR is a closed-form localization estimator with a reduced sensor network. The proposed method is a computationally efficient, two-stage estimator that fuses time-difference-of-arrival (TDOA) and frequency-difference-of-arrival (FDOA) measurements with a minimal number of sensors to localize a moving source in IoT settings.
@article{kazzazi2025clear, title = {{CLEAR}: A Closed-Form Minimal-Sensor {TDOA}/{FDOA} Estimator for Moving-Source {IoT} Localization}, author = {Kazzazi, Mohammad and Morsali, Mohammad and Amiri, Rouhollah}, journal = {arXiv preprint arXiv:2510.04160}, year = {2025}, month = oct, archiveprefix = {arXiv}, primaryclass = {eess.SP}, doi = {10.48550/arXiv.2510.04160}, url = {https://arxiv.org/abs/2510.04160}, }