Mitacs Globalink Research Internship · University of Ottawa, Canada
Supervisor: Dr. Paula Branco

Assessed impact of SMOTE, sample reweighting, and feature engineering on model performance in imbalanced datasets. Automated evaluation across 30+ datasets, identifying generalizable transformation patterns.

• PyTorch
• Scikit-learn
• NumPy

Friedrich-Alexander University (FAU) · Remote
Supervisors: Florian Kordon, Dr.-Ing. Vincent Christlein

Built unsupervised encoder-decoder model to separate bleed-through artifacts from text in degraded historical documents. Created synthetic dataset simulating document degradation for large-scale archival restoration.

• PyTorch
• OpenCV

Sapienza University of Rome · Remote
Supervisor: Dr. Roberto Capobianco

Enhanced classifiers with memory bank and cosine-similarity retrieval to detect spurious shortcuts. Applied memory-guided triplet loss to separate embeddings with shared bias features, reducing shortcut learning.

• PyTorch
• OpenCV