Developing Latent Visual Diffusion Reasoning (LVDR) for interpretable disc-/volume-level reasoning over multi-slice scans (e.g., SPIDER, BraTS, RetinaMNIST), supporting ordinal grading and presence detection.
LVDR builds on a frozen Diffusion Autoencoder encoder and a slot-based memory updated across slices to form a latent reasoning trajectory, enabling step-by-step interpretability via trajectory analyses and diffusion-based reconstructions.
Contributed to Human–Object Interaction (HOI) detection (V-COCO, HICO-DET) by curating a basketball-specific HOI dataset and analyzing QPIC-style transformer baselines.

Supported a foundation-model framework for Cryo-electron Tomography (Cryo-ET), focusing on representation learning under low signal-to-noise conditions.
Worked on equivariant transformers and contrastive learning components for subtomogram alignment and protein particle classification.

Leading a team developing data-efficient computer vision for dermoscopic image analysis to strengthen early skin-cancer detection in underrepresented populations.
Co-founded the Applied Machine Learning Lab, shaping the research roadmap and fostering collaborations through the Brownian Labs Think Tank.

Conducting research under , , and .
Co-authored work on optimization for applied ML, including DBSMOA, Ocotillo Optimization–guided deep learning, and hybrid DTO–PRS models—emphasizing efficient training, feature/model selection, and robust performance across datasets.