Yifan Sun

FedSEPT achieves superior local adaptation without sacrificing global generalization, even under stringent privacy conditions.

RL amplifies existing preferences in LLMs while also revealing previously hidden correct moves, reshaping our understanding of model training dynamics.

By intelligently perturbing class prototypes based on their discriminative power, VPDR achieves a superior privacy-utility trade-off in federated learning compared to naive Gaussian noise.

Frustrated with clunky architecture simulators? Akita offers a breath of fresh air with its focus on developer experience, promising faster prototyping and experimentation.

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.

LLMs' chain-of-thought explanations often fail to reflect the true drivers of their decisions, and this benchmark reveals that closed-source models are particularly opaque, with monitorability dropping by up to 30% under stress.