About Me

I am a second-year MPhil student in the CityMind Lab at the Hong Kong University of Science and Technology (Guangzhou), under the supervision of Dr. Yuxuan Liang and Prof. James T. Kwok. Previously, I obtained B.Eng in Computer Science and Technology from the College of Computer Science, Beijing University of Technology, where I conducted research under the guidance of Prof. Gengyu Lyu at the DMS Lab in the College of Computer Science. During my undergraduate research, I was also privileged to receive guidance from Prof. Yin Liang and Prof. Xiliang Liu from the College of Computer Science. I completed a research internship at ByteDance, focusing on industrial time series forecasting in 2025. It is also my great pleasure to collaborate with members of the CityMind Lab as well as student researchers from outside the lab, including Yiming Huang, Yanyun Wang and Zihao Wang.

My current research interests focus on advancing data-driven deep learning, with a particular emphasis on spatiotemporal modeling and time series analysis for real-world applications.

  1. Spatiotemporal Modeling and Time Series Analysis:
    • Real-World Applications: Traffic management, meteorological forecasting, and neuroscience.
    • Robust Deep Learning Approaches: Capturing complex spatial and temporal dependencies.
    • Reliable Forecasting: Guiding decision-making and uncovering dynamic process evolution.
  2. Multi-Modal Learning and Generative AI:
    • Synergy with Time Series Analysis: Unlocking broader possibilities for real-world deployments.
    • Versatile and Scalable Solutions: Pushing the boundaries of predictive modeling for tangible benefits in areas such as urban planning and neuroscience research.

My Quote

“The purpose of computing is insight, not numbers.”

—— Richard Wesley Hamming

News

  1. One paper about Time Series Forecasting has been accepted by ICASSP'26.

  2. One paper about Time Series Forecasting has been accepted by KDD'26.

  3. Two papers about Irregular Time Series Forecasting and Spatio-Temporal Forecasting have been accepted by AAAI'26.

  4. Invited talk on "Deep Time Series Forecasting with Time-frequency Transformations" at BJUT.

  5. One paper about Point Cloud Salient Object Detection has been accepted by AAAI'25.

  6. A new life at HKUST(GZ) has begun! Feel free to say hi if we cross paths on campus!

  7. I was awarded as Outstanding Graduate Thesis and Outstanding Graduates of Beijing.

  8. One paper about Time Series Forecasting has been accepted by IJCAI'24.

  9. I received the offer from the Red Bird MPhil Program, HKUST(GZ).

  10. One paper about PM2.5 Forecasting has been accepted by Atmosphere.

  11. One paper about ASD Detection has been accepted by PRICAI'23.

Publications and Preprints

* denotes corresponding author · ^ indicates equal contribution
Canonical pre-alignment framework overview

Revitalizing Canonical Pre-Alignment for Irregular Multivariate Time Series Forecasting

Ziyu Zhou, Yiming Huang, Yanyun Wang, Yuankai Wu, James Kwok*, Yuxuan Liang*

The 40th Annual AAAI Conference on Artificial Intelligence (AAAI'26, CCF-A & CORE-A*)

We fully leverage the Canonical Pre-Alignment technique for irregular multivariate time series forecasting, proposing a lightweight framework to process the pre-aligned series.

arXiv 2025
WaveTS multi-order wavelet derivative transform

Multi-Order Wavelet Derivative Transform for Deep Time Series Forecasting

Ziyu Zhou, Jiaxi Hu, Qingsong Wen, James T. Kwok, Yuxuan Liang*

We propose the Wavelet Derivative Transform to model the derivative of time series and integrate it into a multi-branch network (WaveTS) for efficient and effective time series long-term and short-term forecasting.

arXiv 2025
SDformer architecture diagram

SDformer: Transformer with Spectral Filter and Dynamic Attention for Multivariate Time Series Long-term Forecasting

Ziyu Zhou, Gengyu Lyu*, Yiming Huang, Zihao Wang, Ziyu Jia, Zhen Yang

The 33rd International Joint Conference on Artificial Intelligence (IJCAI'24, CCF-A & CORE-A*)

The Only Long Oral Paper of the Time Series Session (1/12)

We propose a novel Transformer architecture (named SDformer) for long-term time series forecasting. It is the first time to address the problem of smooth attention distribution when modeling time series data with a large number of variates.

Paper LinkPDFCodePoster
TimesNet-PM2.5 methodology overview

TimesNet-PM2.5: Interpretable TimesNet for Disentangling Intraperiod and Interperiod Variations in PM2.5 Prediction

Yiming Huang^, Ziyu Zhou^, Zihao Wang^, Xiaoying Zhi, Xiliang Liu*

Atmosphere (JCR-Q3)

In this paper, we accomplish task-specific adaption of TimesNet (ICLR '23) named TimesNet-PM2.5. This specialized version improved the performance and interpretability of the PM2.5 prediction of Haikou, Hainan Province.

Paper Link
STFM spatial and temporal fusion framework

STFM: Enhancing Autism Spectrum Disorder Classification Through Ensemble Learning-Based Fusion of Temporal and Spatial fMRI Patterns

Ziyu Zhou^, Yiming Huang^, Yining Wang^, Yin Liang*

The 20th Pacific Rim International Conference on Artificial Intelligence (PRICAI '23, CCF-C & CORE-B)

We propose a Spatial and Temporal framework named STFM based on cross-attention for better autism spectrum disorder classification. The results show that STFM's classification accuracy surpassed that of many machine learning models.

Paper LinkPDF
CoC-GAN image generation examples

CoC-GAN: Employing Context Cluster for Unveiling a New Pathway in Image Generation

Zihao Wang^, Yiming Huang^, Ziyu Zhou^

We employ Context-Clustering Block (ICLR '23) into GAN for better interpretability.

arXiv 2023

Education Background

2024.09 - 2026.06 (expected)

MPhil in Data Science and Analytics

Hong Kong University of Science and Technology (GZ)

Guangzhou, Nansha

2020.09 - 2024.07

BEng in Computer Science and Technology with Honours Degrees

Beijing University of Technology

Outstanding Graduate Thesis 北京市优秀毕业设计（论文） (Top 0.7%) · Outstanding Graduates 北京市优秀毕业生 (Top 11%) · Beijing, Chaoyang