YZYuandong Zhang

Yuandong Zhang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Yuandong Zhang has a total of 11 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Amplifying Variability of the Southern Annular Mode in the Past and Future

    by Ding Ma, Yuandong Zhang, Ji Nie

    The Southern Annular Mode (SAM) is the dominant mode of atmospheric variability in the Southern Hemisphere extratropics. While its long‐term positive trend is well‐documented, changes in the amplitude of SAM variability remain poorly understood. Based on reanalysis data and surface‐station observations, we demonstrate that the amplitude of SAM variability has substantially increased over the past eight decades, which is associated with more frequent and severe weather extremes. A simple model and idealized climate experiments reveal that this amplification is primarily driven by enhanced stochastic eddy momentum flux, which is linked to a strengthened meridional temperature gradient. State‐of‐the‐art climate model projections indicate that SAM variability will likely continue to intensify throughout the 21st century, driven by a further increase in the meridional temperature gradient. These findings highlight a critical but previously overlooked response of large‐scale circulation to climate change, with important implications for assessing and mitigating future climate risks across the Southern Hemisphere.


  • Open Access

    Federated Black-box Prompt Tuning System for Large Language Models on the Edge

    by Yiming Li, Jingwei Sun, Yudong Liu, Yuandong Zhang, Ang Li, Beidi Chen, Holger R. Roth, Daguang Xu, Tingjun Chen, Yiran Chen

    Federated learning (FL) offers a privacy-preserving way to train models across decentralized data. However, fine-tuning pre-trained language models (PLMs) in FL is challenging due to restricted model parameter access, high computational demands, and communication overheads. Our method treats large language models (LLMs) as black-box inference APIs, optimizing prompts with gradient-free methods. This approach, FedBPT, reduces exchanged variables, boosts communication efficiency, and minimizes computational and memory costs. We demonstrate the practical implementation of FedBPT on resource-limited edge devices, showcasing its ability to efficiently achieve collaborative on-device LLM fine-tuning.