Urban Chemical Dispersion using Physics-Informed Neural Networks
Developed a Physics-Informed Neural Network (PINN) to model
the dispersion of hazardous chemical tracers in urban
environments, solving the Convection-Diffusion Equation.
Achieved an L2 error score of 0.45.
Quantum Siamese Networks for Contrastive Learning
Experimented with different quantum circuits to generate
quantum embeddings of classical data on the MNIST dataset.
Minimized the fidelity between negative pairs (0.55) and
maximized the fidelity between positive pairs (0.90).
Achieved an ROC curve with an excellent AUC of 0.99.
Credit Card Behaviour Score Prediction
Developed a predictive model to assess credit card default
probabilities, using tree-based models and neural networks,
achieving 97% accuracy on validation data.
Evaluated models using metrics for imbalanced datasets,
including G-Mean, MCC, Cohen's Kappa.