GBA School Research · Publications

Preprints and academic outcomes of the Committee and its university partners, focused on AI safety, data governance and trustworthy intelligence.

Preprints

Building the "International TCM Data Port": A Cross-Border Citizen Science Paradigm for the "Data Element ×" Strategy

Liu Kaijie · Alex Li · Hu Hao · Ho Kin Chung

2026 · SSRN

DOI: 10.2139/ssrn.6368638

Scaling Laws Hit a Wall: Mitigating Semantic Sorting Collapse in Medical LLMs via the HoloAuditor Topological Guardrail

Liu Kaijie · Alex Li · Hu Hao · Ho Kin Chung

2026 · SSRN

DOI: 10.2139/ssrn.6368538

When Embodied AI Meets Topology: A Topological Circuit Breaker for OpenClaw and Industrial Agents

Liu Kaijie · Alex Li · Hu Hao · Ho Kin Chung · Michael K. Ng

2026 · SSRN

DOI: 10.2139/ssrn.6421619

HoloTSH: A Neuro-Symbolic Tensor Logic for TCM Modernization via Mathematical Isomorphism and Theoretical Guarantees

Liu Kaijie · Alex Li (Tak Ho) · Hu Hao · Ho Kin Chung

2026-02 · TechRxiv (IEEE)

DOI: 10.36227/techrxiv.177130202.27796764

Intrinsic Risks of the "Survival Motive" in Embodied AI and a Topological Physical Circuit-Breaker (HoloBlocker) Governance Architecture (in Chinese)

2026-04 · ChinaXiv

DOI: 10.12074/202604.00098

Opening Pandora's Box: The Existential Threat of Survival-Driven AI and the Urgent Need for Topological Circuit Breakers

Preprint · SSRN

SSRN 6515000

The "ShenNong AI" Ecosystem: Building a "Digital Great Wall" for Adolescent Physique Data (in Chinese)

ResearchGate

DOI: 10.13140/rg.2.2.26761.81762

A K12 AI-Driven Research Network: A New Paradigm Coupling Data Production and Talent Innovation (in Chinese)

2026-06-11 · ChinaXiv

DOI: 10.12074/202606.00271

HoloAegis: Frozen Representation, Topological Inference — Minimally Parametric Safety Manifolds for Zero-Shot LLM Guardrails

arXiv

DOI: 10.48550/arXiv.2608.08485

Split-Learning and Split-Inference Privacy Defenses Fail Under Strong Attackers: A Metric-Agnostic Dual-Track Empirical Audit of Representation Leakage

2026-09 · ResearchGate

DOI: 10.13140/RG.2.2.32481.67681

Journal papers (under review)

The following 8 papers have been submitted to international journals and are in independent peer review. Until accepted we do not describe them as journal publications; outcomes follow the journals' decisions.

Neural Networks (Elsevier)

2026 · under review

IEEE Transactions on Information Forensics and Security (TIFS)

2026 · under review

Computer Methods and Programs in Biomedicine (CMPB)

2026 · under review

Journal of Systems Architecture (JSA)

2026 · under review

Safety Science (Elsevier)

2026 · under review

IEEE Journal of Biomedical and Health Informatics (JBHI)

2026 · under review

Expert Systems with Applications (ESWA)

2026 · under review

Transactions of the Association for Computational Linguistics (TACL)

2026 · under review

Dataset

HoloAuditor: Topological Guardrail for Clinical LLMs — Audit Ledger, Evaluation Scripts, and Corpus Matching Pairs

2026-10-20 · IEEE DataPort (CC BY 4.0)

DOI: 10.21227/pk5k-cb66　A 1,000-case evaluation ledger, multi-prompt robustness logs and corpus matching pairs with evaluation scripts — fully reproducing the paper's results.

Committee column articles (GDIRI WeChat account)

Reproduced from the Guangdong Digital Industry Research Institute account with the author's permission; each English edition carries a committee editor's note. Click a title to read.

Four universities in GBA launch the AI-safety "Twin Star" architecture

GDIRI · original (Chinese)

Cracking the thousand-year "black box": HoloTSH gives TCM's intuition a mathematical engine

GDIRI · original (Chinese)

A "mathematical brake" for galloping embodied AI

GDIRI · original (Chinese)

Building the Digital Great Wall: topological safety and AI literacy for Hong Kong's future education

GDIRI · original (Chinese)

When AI grows hands and feet: the OpenClaw revolution and a "safety streaking" problem

GDIRI · original (Chinese)