- π MSc in Machine Learning at MBZUAI, advised by Dr. Salem Lahlou and Dr. Martin TakΓ‘Δ.
- π’οΈ Thesis research at AIQ (previously owned by G42 / ADNOC), building time series language models that reason on long horizon industrial sensor data.
- π§ Fatima Fellowship predoctoral researcher with Dr. M. Umar B. Niazi (MIT/KTH) on learned observers for nonlinear control systems.
- π Outside work: boxing, kayaking, and hosting BBQ parties.
π A longer year by year walk through my research journey lives on my academic site: Research Background β
IndusTSLM Time series language models for drilling sensor data. Contrastive dual encoder alignment (DriMM), Flamingo style cross attention, and DrillBench. NeurIPS 2025 Workshop β’ IEEE Big Data 2025 |
HyperKKL Hypernetwork conditioned KKL observers for non autonomous nonlinear systems. 29% SMAPE reduction across four benchmarks under non zero input regimes. ICLR 2026 Workshop β’ CDC 2026 (submitted) |
SVRPBench A realistic benchmark for the Stochastic Vehicle Routing Problem. 500+ instances (10 to 1k customers) with realistic layouts, stochastic delays, and more. NeurIPS 2025 (Datasets & Benchmarks Track) |
Watermark-Analysis Adaptive attacks and evaluation pipeline for invisible image watermarks. π₯ 1st place on both tracks of the NeurIPS 2024 "Erasing the Invisible" challenge. ICLR 2025 Workshop on GenAI Watermarking |
|
πΈ FedPara baseline in Flower
Reproducibility baseline for FedPara (ICLR 2022) contributed to the Flower federated learning framework during the 2023 Summer of Reproducibility at Cambridge. |
π FedSecAgg
Personalized federated neural collaborative filtering with secure multi party computation (SMPC) aggregation, integrated into Flower. Bachelor's thesis, supervised by Dr. Ahmed Kosba. |