Yifan Sun

By intelligently perturbing prototypes based on class variance, VPDR achieves a superior privacy-utility trade-off in personalized federated learning compared to naive Gaussian perturbation.

Stop wrestling with clunky architecture simulators: Akita offers a streamlined, high-usability framework that lets you focus on design, not infrastructure.

Pre-trained foundational models can guide the fusion of optical and SAR features to substantially improve multimodal change detection, achieving state-of-the-art results.