Portrait of Siru Zhong

Siru Zhong

I am a Ph.D. candidate at HKUST (Guangzhou), affiliated with HKUST, advised by Prof. Yuxuan Liang and Prof. James T. Kwok. I am visiting Tsinghua University, hosted by Prof. Haohuan Fu.

I study how machine learning models learn from temporal data that keep evolving, including time series, spatio-temporal signals, and video streams. My work has appeared at venues such as NeurIPS, ICML, KDD, WWW, AAAI, and ACM MM, with several papers selected for oral or spotlight presentation. I have presented tutorials at AAAI and ACM MM. Beyond academia, I have also worked at Tencent, XPENG, Huawei, the National Supercomputing Center, and Beta Infinity.

News

  • I gave an invited talk on memory for video understanding at the Greater Bay Area Graduate Research Forum, CUHK (SZ).
  • BetaMem (long-term memory for embodied agents) was released as a technical report.
  • EarthVerse (benchmarking scientific agents across Earth systems and natural hazards) was released.
  • LiveHouse-TS (an open-world living benchmark for time series foundation models) was released.
  • ViST (multimodal spatio-temporal forecasting) was accepted to ACM MM 2026.
Earlier news
  • I gave an invited talk on spatio-temporal learning at the 6th Young Scholars Forum, Central South University.
  • FARS (a fully automated research system deployed at scale) was released as a technical report.
  • I joined Beta Infinity as a Research Intern working on long-term memory.
  • I presented DropoutTS at AI TIME × ICML 2026.
  • I passed the Ph.D. qualifying exam and became a Ph.D. candidate at HKUST (GZ).
  • TS-Memory (plug-and-play memory for time series foundation models) was accepted to KDD 2026.
  • DropoutTS (robust time series forecasting) was accepted to ICML 2026.
  • I joined Tsinghua University and the National Supercomputing Center as a visiting student, focusing on temporal sequence modeling.
  • I presented a tutorial on multimodal time series analysis at AAAI 2026 in Singapore.
  • I gave a talk on learning across space and time at Tsinghua Shenzhen International Graduate School.
  • OccamVTS (distilling vision models for time series forecasting) was accepted to AAAI 2026.
  • I presented FactoST at the NeurIPS 2025 Symposium, Southern University of Science and Technology.
  • I presented FactoST at AI TIME × ICML 2025.
  • FactoST (factorized spatio-temporal foundation models) was accepted to NeurIPS 2025 as a spotlight.
  • STUM (traffic flow forecasting) was accepted to IEEE TITS.
  • Our paper on multimodal building electricity load forecasting was accepted to ACM MM 2025.
  • DeepUHI (urban heat island forecasting) was accepted to KDD 2025.
  • Time-VLM (vision-language models for time series forecasting) was accepted to ICML 2025.
  • I joined Huawei as a Research Intern working on spatio-temporal foundation models.
  • I started my Ph.D. studies at the Hong Kong University of Science and Technology (Guangzhou).
  • I received the Runner-Up Prize in the 2024 HKUST (GZ) DSA Excellent Research Award.
  • UrbanVLP (urban indicator prediction) and AirRadar (nationwide air quality inference) were accepted to AAAI 2025.
  • I defended my MPhil thesis in Data Science and Analytics at HKUST (GZ).
  • UrbanCross (satellite image-text retrieval) was accepted to ACM MM 2024.
  • SINPA (parking availability prediction) and STFNN (air quality inference) were accepted to IJCAI 2024.
  • I joined XPeng Motors as a Research Intern working on spatio-temporal perception for autonomous driving.
  • UrbanCLIP (LLM-enhanced urban region profiling) was accepted to WWW 2024.
  • I started my MPhil studies at the Hong Kong University of Science and Technology (Guangzhou).
  • I joined Tencent as a Software Engineer, working on a temporal workflow engine.
  • I graduated from Hefei University of Technology with the Outstanding Thesis Award.
  • I completed my three-month internship at Tencent and received a return offer.

Research

* equal contribution Full publication list

My work asks three questions: (i) how to represent temporal dynamics, (ii) how to stay reliable when data shift, and (iii) what to remember from the past.

Representing Dynamics

Multimodal Spatio-Temporal Foundation Models

Adapting to Shift

Robust Learning under Noise and Non-Stationarity

Remembering over Time

Memory for Long-Horizon Understanding

Education

Research & Industry Experience

  • Jun 2026 – Present

    Research Intern, Beta Infinity, Foundation Model Lab · Long-term Memory

  • Feb 2026 – Jun 2026

    Visiting Researcher, NSCS, HPC Department · Temporal Sequence Modeling

  • Feb 2025 – Dec 2025

    Research Intern, Huawei, 2012 Laboratories · Spatio-Temporal Foundation Model

  • May 2024 – Nov 2024

    Research Intern, XPENG, Autonomous Driving Center · Multimodal Perception

  • Jun 2022 – May 2023

    Software Engineer, Tencent, PCG Tech Center · Temporal Data Infrastructure

  • Jun 2021 – Sep 2021

    Software Intern, Tencent, PCG Tech Center · Temporal Workflow Engine

Service

Reviewer
NeurIPS 2026; FMTS 2026 @ NeurIPS 2026; ICLR 2025, 2026, 2027; KDD 2026, 2027; WWW 2027; ACML 2026; MM 2025; IEEE TPAMI; Neurocomputing; IJITDM
PC Member
WSDM 2027; ACM MM 2026; AAAI 2026, 2027; IJCAI 2027; PAKDD 2026, 2027; MILETS 2026 @ KDD 2026
Tutorial Speaker
MM4ST @ MM 2025, MM4ST @ ICME 2026, MM4TS @ AAAI 2026
Web Chair
WebST 2026, WebST 2025, UrbComp 2025

Talks

Tutorials

Invited Talks

Teaching

Awards

  • 2026Junior Student Track Award, DSA Excellent Research Award, HKUST (GZ)
  • 2025Runner-Up Prize for DSA Excellent Research Award, HKUST (GZ)
  • 2023Best Project Award (1st/15) in Data Science Computing, HKUST (GZ)
  • 2023Outstanding Student (Top 10%) in Red Bird Summer Camp, HKUST (GZ)
  • 2022iCode Certification of R&D Engineering Competency Evaluation, Tencent
  • 2022Silver Award (2nd/12) in Code World Program, Tencent
  • 2022Outstanding Student Award (8th/100) in New Employee Training, Tencent
  • 2022Outstanding Graduation Thesis Award (Top 2%), HFUT
  • 2020First Prize (Top 10) in CSDN Technology Blogger Competition