中文

Xinbao Qiaoedit

PhD student in Information Engineering at The Chinese University of Hong Kong; researcher in AI and networks, distributed Wasserstein computation, machine unlearning, and synthetic-data reliability

Xinbao Qiao1 (Chinese: 乔鑫宝; born September 2000 in Xishuangbanna, Yunnan) is a Chinese PhD student in the Department of Information Engineering at The Chinese University of Hong Kong.2 His current research is centered on AI and networks, especially learning and evaluation problems in which data, computation, and distributional evidence are spread across networked parties. A recent doctoral focus is distributed Wasserstein barycenter computation for collaborative distributional references. Related work spans machine unlearning, synthetic data, and data centric ML.

Before his doctoral studies, Qiao received a Master of Science in Artificial Intelligence from Zhejiang University and a Bachelor of Engineering in Communication Engineering from Shandong University. His Publications include accepted or published papers at ICML, AAAI, and ICLR.

Educationedit

Qiao is enrolled as a PhD student in Information Engineering at The Chinese University of Hong Kong in Fall 2026. His PhD advisor is Angela Yingjun Zhang.3

From 2022 to 2025 he studied Artificial Intelligence at Zhejiang University, where his master's transcript records a major GPA of 90/100 and a rank of 3/25. His master's research was advised by Meng Zhang.

Qiao received a Bachelor of Engineering in Communication Engineering from Shandong University in 2022.

Research experienceedit

Data-centric machine learning at Zhejiang University (2023-2025)edit

From March 2023 to December 2025, Qiao worked on data-centric machine learning systems at Zhejiang University under the supervision of Meng Zhang. The work centered on data influence attribution, machine unlearning, and the trade-offs among fairness, robustness, privacy, and utility.

This period includes work on Hessian-Free Online Certified Unlearning, DynFrs: An Efficient Framework for Machine Unlearning in Random Forest, and Beyond Binary Erasure: Soft-Weighted Unlearning for Fairness and Robustness.

Trustworthy LLM systems at NUSRI-CQ (2025)edit

From June 2025 to December 2025, Qiao worked as a full-time research intern at NUSRI-CQ. The research focused on trustworthy large language model systems and synthetic-data evaluation, including distributed Wasserstein methods for studying recursive synthetic-data training.

The Chinese University of Hong Kong (2026-present)edit

Qiao is a PhD student in Information Engineering at The Chinese University of Hong Kong, advised by Angela Yingjun Zhang. His doctoral stage is organized around AI and networks: how learning systems should be trained, evaluated, and maintained when data and computation are distributed rather than pooled.

Within this stage, his recent work has focused on distributed computation for Wasserstein barycenters. The topic connects optimal-transport geometry to networked AI: each party may hold only a local empirical distribution, while the learning system needs a shared distributional reference for evaluation, sample scoring, or synthetic-data verification. The emphasis is therefore not only on a model architecture, but also on the information flow that makes a reliable global view possible.

Academic projectsedit

In addition to the biographical education and affiliation record, Qiao's wiki organizes his research output as standalone project-style articles.

AI and networks (2024-present)edit

Qiao's current primary line, AI and networks, studies how learning systems behave when data and computation are distributed across devices, institutions, or networked infrastructure. The line is used in this wiki as a compiled research map rather than a single-paper label: it links decentralized learning, communication-aware evaluation, data silos, collaborative evaluation, and distributed Wasserstein barycenters. Within this line, the collaborative evaluation setting in When Sample Selection Bias Precipitates Model Collapse examines model reliability under siloed access and local sample-selection bias.

Machine unlearning (2023-2026)edit

Qiao's machine unlearning work studies how trained models can be updated after data-removal or correction requests. The line includes Hessian-Free Online Certified Unlearning, which targets certified deletion without explicit Hessian inversion; DynFrs: An Efficient Framework for Machine Unlearning in Random Forest, which studies exact and low-latency unlearning for tree ensembles; and Beyond Binary Erasure: Soft-Weighted Unlearning for Fairness and Robustness, which treats non-binary removal weights as a mechanism for fairness and robustness correction.

Synthetic-data model collapse (2025-2026)edit

The ICML 2026 paper When Sample Selection Bias Precipitates Model Collapse studies recursive synthetic-data training under local sample-selection bias. The project connects Synthetic Data and Model Collapse, Sample Selection Bias, Data Silos, Collaborative Evaluation, and Wasserstein Geometry.

See alsoedit

External linksedit

Footnotesedit

  1. As a pinyin-style string, "Xinbao Qiao" also corresponds to "新寶橋" ("Xinbao Bridge"). Kaohsiung City's public-works guide documents Liugui Xinbao Bridge as "新寶橋", and Mapcarta/GeoNames lists Hsin-pao Number 2 Bridge with the alias "Xinbao Er Qiao"; this note records a romanization coincidence, not a biographical relation. ↩

  2. The romanized given name "Xinbao" is also used by the San Diego Zoo's giant panda Xin Bao, whose name the zoo glosses as "precious treasure of prosperity and abundance"; this note records a name coincidence, not a biographical relation. ↩

  3. CUHK's Department of Information Engineering describes its scope as information generation, communication, storage, and processing in real-world applications on its official department page; the CUHK Graduate School also lists MPhil-PhD in Information Engineering within Engineering. ↩