Dara VaramMS Machine Learning @ AUS. I did my undergrad here in Computer Engineering and Mathematics with a minor in Data Science. I'm Iranian and I'm 23 years old. I'm currently a visiting student @ MIT Senseable City Lab. I also work as a research & teaching assistant with the Department of Computer Science & Engineering @ AUS. My research focuses mainly on model compression, optimization and quantization for neural networks and LLMs. I have also done studies in training optimization, computer vision, and eXplainable AI. The picture on the right isn't what I look like now. I just wanted to put it up there so you know I've always been cool 👺 b00081313 [at] aus [dot] edu | varam [at] mit [dot] edu LinkedIn | GitHub | ResearchGate | Scholar |
|
Academic Qualifications
Previous: BSc. Computer Engineering, BSc. Mathematics |
American University of Sharjah
Machine Learning Thesis
|
PublicationsThese are papers that have already been published, as reflected on my Google Scholar page (and other venues). |
|
Class Separation Dynamics in Vision Transformers: An Empirical StudyD. Varam, L. Khalil, M. Darwish, M. I. AlHajri We study the class separation dynamics of Vision Transformers (ViTs) as they train for classification tasks, empirically deriving two “laws” that are always followed as a function of training. These two laws have implications that can strongly guide the training strategy and model optimization for transformer-based learning. |
|
Cognitive Radio Spectrum Sensing on the Edge: A Quantization-Aware Deep Learning ApproachH. A. Abushahla, D. Varam, M. I. AlHajri We study the effect of quantization-aware-training (QAT) on two SOTA spectrum sensing models - DeepSense and ParallelCNN. Models are deployed on a Sony Spresense for hardware evaluation. |
|
Image-Scaling Attack Prevention in Industrial Cyber Physical SystemsD. Varam, A. Bayoumy, D. A. Abuhani, M. Zulkernine We propose a novel methodology for detecting and preventing image-scaling attacks for industrial cyber-physical systems using open data. |
|
Estimating Nitrogen Dioxide Levels Using Open Data and Machine Learning: A Comparative Modeling StudyDara Varam, Rohan Mitra, Furzeen Kamran, Diaa A. Abuhani, Hana Sulieman, Imran Zualkernan We study trends in NO2 levels through publicly available geospatial data, specifically in the Lombardy region of Italy. |
|
On-Edge Deployment of Vision Transformers for Medical Diagnostics Using the Kvasir-Capsule DatasetDara Varam, Lujain Khalil, Tamer Shanableh Applying quantization and optimization techniques to a publicly available wireless capsule endoscopy dataset. Paired with a mobile application for deployment. |
|
Variable Selection in Data Analysis: A Synthetic Data ToolkitRohan Mitra, Eyad Ali, D. Varam, Hana Sulieman, Firuz Kamalov Follow-up study on developing synthetic datasets algorithmically for variable / feature selection algorithms. |
|
Wireless Capsule Endoscopy Image Classification: An Explainable AI ApproachDara Varam, Rohan Mitra, Meriam Mkadmi, Radi Aman Riyas, Diaa A. Abuhani, Salam Dhou, and Ayman Alzaatreh We explore eXplainable AI techniques on a Wireless Capsule Endoscopy dataset (publicly available) to better understand model decision-making and classification. |
|
Construction of Multiplicative Groups of Polynomials with Non-Zero Identities in Z_p[x]Dara Varam, Ayman Badawi Explorinig construction of multiplicative groups of polynomials with non-zero identities. |
|
A Hybrid Rolling and Flying Robot for Pipeline Fault Detection Using Deep LearningMegan Ghaly, Rohan Mitra, Abdullah Al Rayess, Assem Ahmed, D. Varam, Mohammad A. Jaradat, Michel Pasquier, Ahmed Khalil Designed, built and programmed a hybrid drone designed for detecting faults on the bodies of oil and gas pipelines. |
|
Lumerical Simulation of Surface-illuminated Silicon PIN Photodiodes for Datacenter InterconnectsS. Ghandiparsi, D. B. Hamadou, D. Varam, A. Soufi, T. Landolsi, A. F. Elrefaie, A. S. Mayet, C. B. Perez, E. P. Devine, S. Y. Wang, T. Yamada, M. S. Islam Studying different characteristics of two novel silicon photodiode (SiPD) devices at small- and large-scale. |
|
Development of Synthetic Data Benchmarks for Evaluating Feature Selection AlgorithmsRohan Mitra, Dara Varam, Eyad Ali, Hana Sulieman, Firuz Kamalov Developing synthetic datasets for benchmarking the performance of feature selection algorithms. First of two works that explores the field. |
Accepted PapersThese are papers that have been accepted for publication, but are not yet up. |
Submitted PapersThese are papers that have been submitted for publication, but have not yet been released. |
|
Quantized Neural Networks for Microcontrollers: A Comprehensive Review of Methods, Platforms, and ApplicationsH. A. Abushahla, D. Varam, A. J. N. Panopio, and M. I. AlHajri We comprehensively review the quantization landscape, specifically for microcontroller-class (MCUs) devices. The survey covers advanced quantization techniques, hardware platforms as part of three families (ARM-based, RISC-V-based and NPU-based), and the software toolchains. |
|
Low-Bit, High-Fidelity: Optimal Transport Quantization for Flow MatchingAnonymous Submission We study the effects of an adapted “Optimal Transport”-based quantization on Flow Matching models, showing better preservations of generating quality and latent disentanglement at extreme low-bits (sub-4 bits). |
|
Exploring Cultural Biases of Arabic-Speaking Large Language Models: A Word Embedding PerspectiveL. Khalil, A. Bayoumy, D. Varam, A. Aklson We study the existence of quantitative bias in Arabic-speaking LLMs. This is done through a word-embedding perspective, specifically looking at the way words are represented in models in terms of sentiment / association and clustering. |
Research ProjectsThis includes other projects that I am currently working on, which have not yet been submitted or published anywhere. |
Presentations, FeaturesIncludes presentations conducted online and in-person, and other miscellaneous academic things I've been featured in. |
|
UAE: How AI is a 'game changer' in academics and industry; educators urge 'honest' useDara Varam I was featured on an article regarding AI and its uses in academia. Feel free to take a read based on the website provided. |
|
eXplainable AI for Image Classification Tasks in MedicineDara Varam Spoke at the Google I/O Extended UAE 2024 keynote, organized through Google Developer Groups Sharjah. |
Miscellanea |
Peer Reviewed Articles
- Association for the Advancement of Artificial Intelligence (AAAI) - 2026 Conference – 4 papers
- Nature – Scientific Reports – 4 papers
- Nature – BioData Mining – 2 papers
- Nature – Discover Computing – 1 paper
- Nature – Signal, Image and Video Processing – 1 paper
- Nature – Cluster Computing – 1 paper
- Nature – Applied Sciences – 1 paper
Teaching Assistantships
- Graduate-level: MLR570 - Advanced Machine Learning
- Software: CMP120 / CMP220 / CMP305 / CMP320 – Programming I & II, Data Structures & Algorithms, Database Systems
- Hardware: COE242 / COE251 – Microcontrollers & Embedded Systems, Intro. to Computer Systems
- Networks: COE271 – Computer Networks I
- Mathematics: MTH103 / MTH104 / MTH203 / MTH205 / MTH221 – Calculus I, II & III, Differential Equations, Linear Algebra
Awards & Honors
- 1st place at the AI-City Quest Hackathon.
- 2nd place at the Planet-X Mars Rover Mission challenge.
- Member of the IEEE-Eta Kappu Nu Honors Society.
- Member of the AUS Engineering Honors Society.