• 🎓 Dual Master's Degrees in AI/Data Science (Dauphine–ENS–Mines Paris) & Computational Modeling (Université Côte d'Azur)
• 🧠 Experienced in Machine Learning, reinforcement learning, and multi-modal modeling
• 🌍 International experience in France 🇫🇷, Switzerland 🇨🇭, and the UAE 🇦🇪
• 💬 Multilingual: French, English, Arabic, Turkish
• 📊 Skilled in Python, ML, DL

🔭 Applying data science to real-world problems 📚 Building reproducible ML pipelines for real-world data 📝 Contributing to open-source & scientific publications

Languages & Tools:

Python, R, SQL, Java, C | Scikit-learn, TensorFlow, Keras, PyTorch
Pandas, NumPy, Matplotlib, Seaborn | Git, GitLab, Linux, Jupyter, GCP

Domains:

• Machine Learning & Deep Learning
• Biomedical Data Analysis (EEG, MRI, health records)
• Reinforcement Learning, Multi-agent systems
• Data Visualization & Explainability

Tools: Scikit-Learn · SHAP · Pandas · Matplotlib
Tags: Classification · Healthcare · Explainability · Portfolio

End-to-end ML pipeline using the UCI Heart dataset with:

• EDA, model tuning (LogReg, RF, KNN), cross-validation
• SHAP-based model explainability for clinical interpretation
• Clear markdown structure, visual summaries, and performance metrics

📌 View Project on GitHub

Tools: NumPy · Scikit-Learn · Matrix Factorization · SVD · KNN
Tags: Recommender System · Sparse Data · Ensemble Learning

Tackled extreme data sparsity using:

• Matrix Factorization + Gradient Descent
• k-NN with cosine similarity
• SVD for low-rank approximations
• A custom ensemble for robust final predictions

📌 Explore the System on GitHub

Tools: PyTorch · NumPy · F-divergences · GANs
Tags: Generative Modeling · Rejection Sampling · Deep Learning

Explored advanced GANs through:

• Training f-GANs with different f-divergence losses (JS, KLD, BCE)
• Implementing Discriminator Rejection Sampling for improved generation
• Evaluating models using FID, precision, and recall

📌 View the Full Project

Training Robust Neural Networks against Adversarial Attacks

Tools: PyTorch · Adversarial ML · CIFAR-10
Tags: CNN · Robustness · Adversarial Training · Explainability

Investigated the robustness of CNNs to adversarial attacks:

• Built baseline CNN models (LeNet-style and improved architectures)
• Implemented FGSM, PGD, and DeepFool adversarial attack methods
• Applied adversarial training and randomized network defenses
• Analyzed performance degradation and recovery under adversarial stress
• Packaged the experiments with clear Jupyter notebooks and scripts

📌 Explore the Project on GitHub

End-to-End Regression Pipeline with Feature Engineering & Random Forests

Tools: Scikit-Learn · Pandas · Random Forest · Tabular ML
Tags: Regression · Price Prediction · Feature Importance · Time-aware Split

Built a full machine learning pipeline to predict the auction price of bulldozers:

• Cleaned and preprocessed 400,000+ auction records from the Blue Book dataset
• Engineered time-based features (saleYear, machineAge) and encoded categoricals
• Trained and validated Random Forest models using RMSLE
• Ranked top contributing features like YearMade, ProductSize, saleYear
• Included a fully documented Jupyter notebook with EDA, modeling, and evaluation

📌 View the Project on GitHub

Simulated human-like movement adaptation using MuJoCo + reinforcement learning.

Transformer-based model achieving 97% accuracy on unimodal EEG decoding. (Will be shared once published)

Multi-level classification system with handcrafted statistical features & ML.

• Machine Learning and Data Science - Zero To Mastery Academy

📫 Email: lynabouiknia@gmail.com
🔗 LinkedIn
📁 Portfolio

Always curious. Always learning. Always building.