Projects
Explore my key research projects and technical implementations in Computer Vision, Medical Imaging, and Federated Learning.
Featured Projects
FedUniBrain: Universal Brain MRI Segmentation
A groundbreaking federated learning framework that enables training a single model to segment multiple types of brain lesions from databases with different MRI modalities, without any data exchange between institutions. This work demonstrates the feasibility of decentralized collaborative learning in medical imaging.
Key Achievements:
- First framework to handle multiple brain diseases and MRI modalities in federated setting
- Accepted at WACV 2025 with oral presentation
- Addresses critical privacy concerns in medical AI
- Enables cross-institutional collaboration without data sharing
StarAlign: Gradient Alignment for Domain Adaptation
An innovative algorithm that enables deployed models to adapt to new target distributions by accessing training data through federated learning, without compromising data privacy. The method aligns gradients between source and target data for effective post-deployment adaptation.
Technical Innovation:
- Novel gradient alignment technique for domain adaptation
- Federated learning approach preserves data privacy
- Addresses distribution shift in real-world deployments
- Presented at MLMI 2023 workshop at MICCAI
F³OCUS: Federated Foundation Model Finetuning
A novel layer updating strategy for parameter-efficient finetuning of Vision-Language Foundation Models in federated settings. F³OCUS optimizes client-specific layer importance and inter-client layer diversity using multi-objective meta-heuristics.
Research Impact:
- Accepted at CVPR 2025 - top-tier computer vision conference
- Novel approach to federated foundation model training
- Addresses resource constraints in federated learning
- Advances parameter-efficient fine-tuning techniques
Research Areas
🧠 Medical Imaging AI
Developing robust AI systems for medical image analysis, with focus on brain MRI segmentation, cross-modal learning, and handling distribution shifts in clinical settings.
🔒 Privacy-Preserving ML
Creating federated learning frameworks that enable collaborative model training across institutions while maintaining strict data privacy and security requirements.
🔄 Domain Adaptation
Researching methods to adapt AI models to new environments and data distributions, ensuring reliable performance in real-world deployment scenarios.
All code implementations follow best practices in software engineering with comprehensive documentation, testing, and reproducible research standards.