Work
Projects
01
Charge Density Prediction
OpenGPNN predicts electron charge density distributions across materials using neural networks trained on atomic structure data. By replacing computationally expensive density functional theory (DFT) calculations with learned models, it achieves predictions orders of magnitude faster — enabling rapid screening of material properties at the atomic scale.
03
3D Brain Tumor Segmentation
A hybrid quantum-classical self-supervised learning architecture modeled after SimCLR, augmented with a variational quantum circuit (VQC) in the projector head of each contrastive learning arm. Using a 3D CNN encoder and an MLP/VQC projector, this proof of concept explored whether quantum ansatze could improve learned representations for volumetric MRI segmentation. Presented at the HIMSS international health conference. Downstream performance did not improve with the VQC while latency increased, highlighting the current limitations of NISQ methods for contrastive learning.
04
Skin Cancer Classification
A computer vision classifier using MobileNet fine-tuned with the fastai framework, achieving ~90% accuracy on skin lesion classification. Demonstrates effective application of transfer learning to high-stakes biomedical imaging, making powerful vision models accessible for clinical screening tasks.