Peiyuan Liu (劉培源)

master@THU

me.jpg

I am currently pursuing my master’s degree in Computer Science at the esteemed Tsinghua University, under the mentorship of Prof. Shu-Tao Xia. I work closely with Prof. Tao Dai. Before joining Tsinghua, I completed my undergraduate studies in Computer Engineering at Zhejiang University and the University of Illinois at Urbana-Champaign.

Throughout my academic journey, I’ve had the privilege of bolstering my classroom knowledge with practical industry experience at Oneflow and Xorbits, startups leading innovation in their respective fields. These experiences have provided me with a unique vantage point to tackle real-world challenges, thereby augmenting my understanding of the theoretical aspects and practical implementations of technology.

My research interests include large language models (LLMs) and long-term time-series forecasting (LTSF). I am driven by the idea of utilizing these sophisticated computational methodologies to unearth deeper insights and design solutions that can have a transformative impact on society. I am committed to pushing the boundaries of these research areas and exploring the full potential of machine learning algorithms to contribute meaningfully to the world of technology.

News

Sep 26, 2024 Our paper DDN: Dual-domain Dynamic Normalization for Non-stationary Time Series Forecasting is accepted by NeurIPS 2024.
Jan 15, 2024 Our paper Periodicity Decoupling Framework for Long-term Series Forecasting is accepted by ICLR 2024.
Dec 14, 2023 Our paper WFTNet: Exploiting Global and Local Periodicity in Long-term Time Series Forecasting is accepted by ICASSP 2024.
Sep 26, 2023 Our paper DIVOTrack: A Novel Dataset and Baseline Method for Cross-View Multi-Object Tracking in DIVerse Open Scenes is accepted by IJCV 2023.
Sep 1, 2023 Become a graduate student at Tsinghua University, advised by Tao Dai and Shu-Tao Xia.
Jun 30, 2023 🎉 Graduate from Zhejiang University and University of Illinois Urbana-Champaign.
May 8, 2023 Join Hangzhou Xorbits Tech Co., Ltd as an Intern in Open Source System Developer.
Feb 2, 2023 Join Beijing Oneflow Tech Co., Ltd as an Intern in Deep Learning Framework Development Engineering.

Publications

\(^*\) Equal contribution. \(^\dagger\) Corresponding author.

2024

  1. NeurIPS
    DDN: Dual-domain Dynamic Normalization for Non-stationary Time Series Forecasting
    Tao Dai, Beiliang Wu, Peiyuan Liu\(^\dagger\), Naiqi Li, Xue Yuerong, Shu-Tao Xia, and Zexuan Zhu
    2024
  2. ICLR
    Periodicity Decoupling Framework for Long-term Series Forecasting
    Tao Dai, Beiliang Wu, Peiyuan Liu\(^\dagger\), Naiqi Li, Jigang Bao, Yong Jiang, and Shu-Tao Xia
    International Conference on Learning Representations, 2024
    Code NPM
  3. ARXIV
    TimeBridge: Non-Stationarity Matters for Long-term Time Series Forecasting
    Peiyuan Liu, Beiliang Wu, Yifan Hu Naiqi Li, Tao Dai, Jigang Bao, and Shu-Tao Xia
    2024
    arXiv Code NPM
  4. ARXIV
    CALF: Aligning LLMs for Time Series Forecasting via Cross-modal Fine-Tuning
    Peiyuan Liu\(^*\), Hang Guo\(^*\), Tao Dai\(^\dagger\), Naiqi Li\(^\dagger\), Jigang Bao, Xudong Ren, Yong Jiang, and Shu-Tao Xia
    arXiv preprint arXiv:2403.07300, 2024
    arXiv Code NPM
  5. ARXIV
    Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting
    Yifan Hu\(^*\), Peiyuan Liu\(^*\), Peng Zhu, Dawei Cheng\(^\dagger\), and Tao Dai
    arXiv preprint arXiv:2406.03751, 2024
    arXiv Code NPM
  6. ICASSP
    WFTNet: Exploiting Global and Local Periodicity in Long-term Time Series Forecasting
    Peiyuan Liu, Beiliang Wu, Naiqi Li, Tao Dai, Fengmao Lei, Jigang Bao, Yong Jiang, and Shu-Tao Xia
    IEEE International Conference on Acoustics, Speech and Signal Processing, 2024
    arXiv Code NPM

2023

  1. IJCV
    DIVOTrack: A Novel Dataset and Baseline Method for Cross-View Multi-Object Tracking in DIVerse Open Scenes
    Shengyu Hao\(^*\), Peiyuan Liu\(^*\), Yibing Zhan, Kaixun Jin, Zuozhu Liu, Mingli Song, Jenq-Neng Hwang, and Gaoang Wang\(^\dagger\)
    International Journal of Computer Vision, 2023
    arXiv Code NPM