Marawan Gamal

Marawan Gamal

PhD Computer Science • Mila / UdeM

Menu

About

I am currently doing my PhD in Computer Science at Mila (University of Montreal) supervised by Prof. Guillaume Rabusseau. Previously I completed my MASc at the Vision and Image Processing lab at University of Waterloo under the supervision of Prof. David Clausi and Prof. John Zelek. I completed my undergraduate degree at the University of Toronto where I specialized in Mechatronics and Robotics Engineering.

My research interests currently are at intersection of Natural Language Processing and Tensor Networks. Previously, I worked in computer vision on video action recognition and as well as model compression of large language models.

I also serve as a Scientist in Residence (SiR) at NextAI, where I provide technical consulting to startups.

Publications

Tensor Decomposition Paper
A Tensor Decomposition Perspective on Second-order RNNs
Maude Lizaire, Michael Rizvi-Martel, M. Gamal Abdel Hameed and G. Rabusseau. International Conference on Machine Learning (ICML) 2024
ROSA Paper
ROSA: Random Subspace Adaptation for Efficient Fine-Tuning
M. Gamal Abdel Hameed, Aristides Milios, Siva Reddy and G. Rabusseau. ICML 2023 Workshop on Efficient Systems for Foundation Models
😔
No demo
Generalized Kronecker-based Adapters for Parameter-efficient Fine-tuning of Vision Transformers
A. Edalati, M. Gamal Abdel Hameed and A. Mosleh. Conference on Computer and Robot Vision (CRV) 2023
CNN Compression Paper
Convolutional Neural Network Compression through Generalized Kronecker Product Decomposition
M. Gamal Abdel Hameed, M. S. Tahaei, A. Mosleh, and V. Partovi Nia. AAAI 2022
SeKron Paper
SeKron: A Decomposition Method Supporting Many Factorization Structures
M. Gamal Abdel Hameed, M. S. Tahaei, A. Mosleh, and V. Partovi Nia. arXiv preprint arXiv:2210.06299, 2022

Special Projects

Dekki.ai and Papers.app are two apps I built during my PhD.

Dekki
50k+ users

Developed a statistical spaced-repetition app for medical students, in collaboration with Dr. Luke Kyne and Prof. Oliver Hardt. The algorithm models each student's retention curve using Ebbinghaus' forgetting model, enabling the optimization of user spaced-repetition schedules. The app grew to over 50k users.

dekki.ai
Live Demo

I wanted to semantically search papers, with options to filter by conference. I built this app in collaboration with Jeremy Pinto.

papers.app
Live Demo
YouTube
100k+ views

Created an explanation of the Singular Value Decomposition (SVD) by animating key concepts from Gilbert Strang's MIT lecture, essentially visualizing the geometric intuition behind SVD.

youtube.com

Contact

Email me at marawan [dot] gamal [at] mila [dot] quebec for research collaborations, speaking opportunities, or consulting inquiries.