Projects
Selected work across machine learning, computer vision, scientific computing, and probabilistic modeling. Each project includes the implementation and technical documentation.
Featured · Computer Vision · Microscopy
A physics-informed complex-valued U-Net that reconstructs high-resolution amplitude and phase images from low-resolution Fourier Ptychographic Microscopy measurements, replacing iterative phase retrieval with a model that learns the reconstruction directly.
PyTorch Physics-informed DL
Computer Vision
A lightweight (under 5M parameters) model that predicts ordered polygon corner coordinates from 128×128 grayscale images. Full training pipeline, variable-length dataset loader, and polygon-IoU evaluation.
PyTorch
Quantum ML · Chemistry
Predicts coupling energies for molecule-pair geometries using classical Kernel Ridge Regression on Coulomb-matrix features, benchmarked against a quantum kernel baseline built with PennyLane.
scikit-learn PennyLane
Computer Vision · Microscopy
Self-supervised denoising of microscopy images using a blind-spot U-Net (Noise2Void) that learns to denoise from a single noisy image, no clean ground truth required.
TensorFlow Noise2Void
Featured · Bayesian Inference
Solves an ill-posed inverse problem analogous to paleoclimate reconstruction, recovering an unknown boundary condition from noisy interior measurements, with evidence-based uncertainty quantification.
NumPy SciPy
Data Augmentation
Application of AugMix, a data augmentation method for improving model robustness and uncertainty calibration.
Python PyTorch