Mohammed ali El adlouni

- Implemented and benchmarked state-of-the-art audio encoders for large-scale foundation model pretraining, leveraging multi-GPU (NVIDIA A100) distributed training for high-throughput experiments. - Built and optimized high-throughput data workflows for the AudioSet dataset (˜2M samples, 650–700 GB). - Designed a self-supervised distillation framework in PyTorch & Lightning, enabling low-latency inference on constrained hardware. - Delivered the most efficient tiny audio foundation model to date, achieving 61× model compression with 96.4% performance retention. - First-author submission to ICASSP 2026: “S-SONDO: Self-Supervised Knowledge Distillation for General Audio Foundation Models.”

Worked on Machine Learning algorithms for Macroscopic Fundamental Diagrams and emission rate predictions based on traffic density, supervised by Dr. Ling JIN. I carried out the following key tasks: - Processed and analyzed a 272M-row spatio-temporal dataset of traffic density, network characteristics, vehicle types, and network-level emissions across U.S. urban census tracts (New York, Colorado, Texas, Georgia). Leveraged multiprocessing on 60 CPU cores to accelerate data cleaning and preprocessing. - Built interpretable predictive pipelines in Python using a benchmark of ML models including Light GBM, CatBoost, XGBoost, and neural networks. - Applied SHAP-based explainability (game-theory principles) to quantify feature contributions. - Developed robust automated testing, validation, and feature engineering workflows, ensuring reproducibility and statistical rigor at scale. - Awarded a $10,000 Research Fellowship. - First-author paper published at IEEE Big Data 2024.

- Developed a real-time localization system for an autonomous EV charging rover using Python, integrating ML models with template matching algorithms in OpenCV for high-precision navigation. - Conducted on-field data collection of 100+ samples. Applied data preprocessing and augmentation strategies. - Designed and implemented end-to-end data pipelines and ML workflows, achieving 95% accuracy in embedded deployment under constrained computational resources.

Practical internship conducted at Stellantis Automotive Group's Sochaux site, involving hands-on experience within the production line, gaining insights into the various stages of production.