Research
My research spans ML, DL, NLP, and computer vision — with a current focus on Neurosymbolic AI, sparse autoencoders and mechanistic interpretability, and federated learning for medical imaging.
Publications
2026
·SN Computer Science (Springer Nature) · Vol. 7, Art. 360 · Q1
DEXNet: An Ensemble Model Integrating DenseNet, EfficientNetB3, and XGBoost for Histopathological Lung and Colon Cancer Classification
Ensemble architecture combining DenseNet and EfficientNetB3 for hierarchical feature extraction with XGBoost classification, augmented by Class-Selective Image Preprocessing (CSIP), Grad-CAM interpretability, and a HIPAA/GDPR-compliant AWS deployment pipeline with federated learning provisions for multi-institutional privacy-preserving inference.
2026
·Discover Computing (Springer Nature) · Vol. 29, Art. 585 · Q2
FedBound: A Boundary-Aware Optimization Strategy for Federated Medical Image Segmentation Under Non-IID Data
A lightweight boundary-aware optimization strategy for federated medical image segmentation that emphasizes contour pixels during local training without adding communication overhead. Evaluated across six segmentation architectures on the ISIC 2018 skin-lesion dataset under a Dirichlet non-IID partition of 100 federated clients — improving boundary precision (HD95) and reducing cross-client variance while preserving Dice/IoU performance.
2024
·ICDEC 2024 · Springer Nature
Barbell Exercise Classification and Repetition Counting
Engineered a robust ML system for barbell exercise classification and repetition counting using MetaMotion sensor data, achieving over 90% accuracy through comprehensive feature engineering and outlier detection pipelines for precise human activity recognition.