Mecharbat Lotfi Abdelkrim

Experience

October 2023 - Now

New York University Abu Dhabi

Research Engineer - Prof. Tuka Elhanai

Domains: AutoML, Quantum Computing, Deep Learning

Project: Enhance Fidelity, Efficiency, and interpretability of deep learning models used for learning quantum systems.

Keywords: Physics Informed Neural Networks, Quantum computing, Neural Architecture Search

Conduct an analysis study comparing the performance of classical and ML-based methods for learning quantum systems.
Based on the insights gained, define the architecture search space that will be fed to the NAS method to enhance the targeted metrics.
Investigate the potential of extrapolating information from 2-qubit systems to simulate and predict the dynamics of 3-qubit systems.

October 2022 - September 2023

ALSTOM / LAMIH

Research Engineer - Prof. Smail Niar

Domains: Efficient Deep Learning, Edge Computing

Project: Diffusion-based Neural Architecture Search for Enhanced Efficiency and Effectiveness

Role: Lead Author

Keywords: Neural Architecture Search (NAS), Diffusion Models, Edge Computing

Developed DiffNAS[4], a diffusion model-based NAS methodology to improve the efficiency and effectiveness of deep learning neural network design.
Validated results on NAS-Bench-101, achieving the highest average accuracy and being at least 8x faster than state-of-the-art methods.

Project: Intelligent Railway Equipment Monitoring with Advanced Anomaly Detection

Role: Project Lead

Keywords: Anomaly Detection, Computer Vision, Self-supervised Learning

Developed a sensor-based system (using onboard cameras) for detecting and diagnosing anomalies in railway equipment using a two-step approach: localization of target objects (e.g., windows, lamps), and anomaly detection within these objects using deep learning models.
Developed a new anomaly generation technique[5] based on diffusion models to construct more effective anomaly detectors.
Optimized resource usage to minimize overhead, ensuring compatibility with other monitoring applications.

Project: Med-nas-bench: A generalized neural architecture search benchmark for medical imaging analysis

Role: Second Author

Keywords: NAS, Medical Imaging, Benchmark

Contributed to the collection and analysis of hardware metrics for the benchmark and assisted with the design of the search space (supernetwork)[3].

September 2021 - September 2022

LAMIH

Research Intern

Domains: Neural rchitecture Search, Computer Vision, Edge Computing, Transformers

Project: Hardware-aware Neural Architecture Search for Efficient Hybrid Architectures on Tiny Devices.

Keywords: NAS, Vision Transformers, Edge Computing, Computer Vision

Conducted a comparative analysis on the performance of Vision Transformers on various hardware platforms.
Proposed HyT-NAS, an efficient HW-NAS for hybrid architectures targeting vision tasks on tiny devices.
HyT-NAS optimized search efficiency by 5x and resulted in models that outperform SOTA on visual tasks with improved accuracy and fewer parameters.

2020 - 2021

AITECH

Data Scientist - Mabrouk Aib

Project: Technical Itinerary Recommendation System for Citrus Production.

Keywords: Optimization, Recommendation System, Machine Learning, Agriculture

Developed a machine learning model to predict citrus crop production quantity based on weather conditions and soil properties.
Constructed a recommendation system for the technical itinerary of citrus production to optimize resource usage and maximize profit.

Publications

DiffNAS: A Diffusion-based Neural Architecture Search

Mecharbat Lotfi Abdelkrim, Hadjer Benmeziane, Kaouther El Maghraoui ,Hamza Ouarnoughi, Smail Niar

SIGMETRICS 2024 (Under review)

Submitted in 10/10/2023

we introduce DiffNAS, a novel NAS methodology rooted in diffusion processes. DiffNAS brings substantial improvements in both NAS search efficiency and the quality of the generated neural network architectures. To the best of our knowledge, our work marks a pioneering effort in applying diffusion algorithms to enhance the search space exploration with NAS. Our experimental results, conducted on the widely-used NAS-Bench-101, showcase the remarkable capabili- ties of DiffNAS. We achieved the highest average accuracy, outperforming other state-of-the-art methods, while completing the search process at least 2 times faster.

Paper Presentation

HyT-NAS: Hybrid Transformers Neural Architecture Search for Edge Devices

Mecharbat Lotfi Abdelkrim, Hadjer Benmeziane, Smail Niar, Hamza Ouarnoughi

ESWEEK 2022

13/10/2022

We developed HyT-NAS, an efficient Hardware-aware Neural Architecture Search (HW-NAS) including hybrid architectures targeting vision tasks on tiny devices. HyT-NAS improves state-of-the-art HW-NAS by enriching the search space, enhancing the search strategy as well as the performances predictors.

Paper Presentation

Certifications

HCIA-Artificial Intelligence:

I am a Huawei Certified ICT Associate in Artificial Intelligence after completing the exam that covers the different techniques of artificial intelligence, especially machine learning, artificial neural networks and Deep Learning.

Data science Track

I completed Datacamp's datascience track that contains more than 20 courses that cover all theoretical and practical aspects in machine learning and its application on different domains such as Computer vision and Natural Language processing.

Data Science Badge and orientation by IBM.

Problem Solving Certificate by HackerRank.

Projects

All
AutoML
Computer Vision
optimization
General Deep Learning
Data Science
Transformers
Edge Computing