Sirui Han
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Sirui Han has a total of 157 co-authors. Here are the top co-authors by number of shared publications:
- Yike Guo
Department of Computer Science and Engineering, Hong Kong University of Science and Technology, Hong Kong, China
22 shared publications - Ruoxi Li
Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea
8 shared publications - Zhizhuo Kou
HKUST, Hong Kong
6 shared publications - Yuyao Zhang
HKUST, Hong Kong
5 shared publications - Chengzhong Liu
The Hong Kong University of Science and Technology, New Territories, Hong Kong
3 shared publications - Guoying Lu
The Hong Kong University of Science and Technology (HKUST)
3 shared publications
Latest Publications
- Open Access
- 11 November 2026
- Published by Elsevier in SSRN Electronic Journal
Is the Model Anonymous? Why Generative AI Forces a Deployment-Based Answer under the GDPR
Debates under the General Data Protection Regulation (GDPR) often ask a deceptively simple question: is the model anonymous? For generative models deployed across different settings, the answer can change with deployment. The same trained model may later be used through a different interface, by another organisation, or with newly connected data, while the testing and documentation supporting an earlier anonymity conclusion remain unchanged. Evidence can therefore remain technically valid yet become insufficient to support the same conclusion for a later processing operation. This article asks how anonymity should be substantiated for a defined operation, how earlier evidence may continue to support that conclusion as deployment conditions change, and how credible contrary evidence should affect the assessment. Drawing on Article 4(1), Recital 26, Court of Justice of the European Union case law, European Data Protection Board Opinion 28/2024, the Board's 2026 consultation Guidelines on Anonymisation, and model-privacy research, it develops two linked propositions. Deployment-Bounded Anonymisation (DBA) identifies the concrete processing operation for which anonymity is claimed as the unit of proof. Conditional evidentiary reuse explains how earlier evidence carries forward: evidence remains probative for propositions whose material conditions remain stable; changes that create or materially strengthen an identification route call for supplementary evidence directed to that route; and credible contrary evidence calls for reconsideration of the proposition it undermines. The durability of an anonymity conclusion therefore depends on whether its supporting evidence remains applicable to the operation in which that conclusion is invoked.
- 11 November 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessIs the Model Anonymous? Why Generative AI Forces a Deployment-Based Answer under the GDPR
Debates under the General Data Protection Regulation (GDPR) often ask a deceptively simple question: is the model anonymous? For generative models deployed across different settings, the answer can change with deployment. The same trained model may later be used through a different interface, by another organisation, or with newly connected data, while the testing and documentation supporting an earlier anonymity conclusion remain unchanged. Evidence can therefore remain technically valid yet become insufficient to support the same conclusion for a later processing operation. This article asks how anonymity should be substantiated for a defined operation, how earlier evidence may continue to support that conclusion as deployment conditions change, and how credible contrary evidence should affect the assessment. Drawing on Article 4(1), Recital 26, Court of Justice of the European Union case law, European Data Protection Board Opinion 28/2024, the Board's 2026 consultation Guidelines on Anonymisation, and model-privacy research, it develops two linked propositions. Deployment-Bounded Anonymisation (DBA) identifies the concrete processing operation for which anonymity is claimed as the unit of proof. Conditional evidentiary reuse explains how earlier evidence carries forward: evidence remains probative for propositions whose material conditions remain stable; changes that create or materially strengthen an identification route call for supplementary evidence directed to that route; and credible contrary evidence calls for reconsideration of the proposition it undermines. The durability of an anonymity conclusion therefore depends on whether its supporting evidence remains applicable to the operation in which that conclusion is invoked.
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 12631-12640
TwinRL: Digital Twin–Driven Reinforcement Learning for Real-World Robotic Manipulation
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 12631-12640
Open AccessTwinRL: Digital Twin–Driven Reinforcement Learning for Real-World Robotic Manipulation
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 7089-7098
Dream-Tac: A Unified Tactile World Action Model for Contact-Rich Robot Manipulation
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 7089-7098
Open AccessDream-Tac: A Unified Tactile World Action Model for Contact-Rich Robot Manipulation
- 3 November 2026
- Published by Taylor & Francis in Journal of Asian Public Policy
- p. 1-25
Ethical disclosure asymmetries in generative AI research: a dual-perspective analysis through national innovation systems and division of labor
1. The rapid expansion of generative artificial intelligence (GAI), particularly systems powered by large language models (LLMs), has heightened concerns over their ethical and societal implication...
- 3 November 2026
- Published by Taylor & Francis in Journal of Asian Public Policy
- p. 1-25
Ethical disclosure asymmetries in generative AI research: a dual-perspective analysis through national innovation systems and division of labor
1. The rapid expansion of generative artificial intelligence (GAI), particularly systems powered by large language models (LLMs), has heightened concerns over their ethical and societal implication...
- Open Access
- 13 October 2026
- Published by Springer Nature in Journal of Orthopaedic Surgery and Research
Age- and sex-specific patterns, seasonality and temporal trends of acute pediatric fracture admissions in China: a pooled multicenter hospital-based cohort of 6206 children
Background: Childhood fractures are a major source of pediatric morbidity and hospital use, yet contemporary large-scale Chinese data with denominator-based modeling and age- and sex-specific analysis remain limited. We characterized the anatomical, seasonal and developmental epidemiology of acute pediatric fracture admissions in a pooled hospital-based cohort. Methods: In this retrospective study, children (< 18 years) with a confirmed acute fracture admitted to three Chinese hospitals between October 2020 and November 2024 were identified from pooled electronic medical records by text mining of structured diagnoses; pathological and old fractures were excluded. Descriptive and regression analyses were patient-level (earliest acute admission per child); seasonality and secular trend in the fracture-admission rate, with all pediatric admissions as the denominator, were modeled at the episode level with cosinor negative-binomial regression. Multivariable logistic regression examined factors associated with multiple-region fracture and with lower-only (vs. upper-only) limb fracture, including the age-by-sex interaction and sex-specific site predilection. Results: Among 251,385 pediatric admissions, 7,142 (2.84%) were acute-fracture admissions in 6,206 children (70.9% male; male: female 2.44:1; mean age 9.7 ± 4.8 years). The upper limb was most often involved (48.9%), led by the humerus (21.6%) and forearm (15.6%); supracondylar humeral fractures comprised 9.7%. The fracture-admission rate showed autumn-peaking seasonality (negative-binomial peak-to-trough ratio 1.31, 95% CI 1.14–1.55; peak in mid-October; p = 0.007) and a declining fracture-admission rate relative to all pediatric admissions (rate ratio 0.82 per year, 95% CI 0.78–0.86). Lower-only (vs. upper-only) limb fracture rose with age (12–17 vs. 6–11 years: OR 2.30, 95% CI 2.00–2.65) and was less frequent in boys (OR 0.58, 95% CI 0.50–0.66); the sex effect varied with age (interaction p < 0.001). Site predilection was sex-specific (boys: forearm, hand; girls: ankle, foot, spine, pelvis, patella, tibia/fibula). Estimates were robust to the unit of analysis, limb coding, data-source stratum and denominator composition. Conclusions: In this hospital-based cohort, acute pediatric fracture admissions showed male and upper-limb predominance, an autumn peak, an age-dependent shift toward lower-limb injury, sex-specific site predilection and a declining share of pediatric admissions over time. Because injury mechanism, activity and a defined source population were unavailable, these patterns describe the hospital admission burden and generate hypotheses for age-, sex- and season-informed prevention research and trauma-service planning rather than establishing specific interventions.
- 13 October 2026
- Published by Springer Nature in Journal of Orthopaedic Surgery and Research
Open AccessAge- and sex-specific patterns, seasonality and temporal trends of acute pediatric fracture admissions in China: a pooled multicenter hospital-based cohort of 6206 children
Background: Childhood fractures are a major source of pediatric morbidity and hospital use, yet contemporary large-scale Chinese data with denominator-based modeling and age- and sex-specific analysis remain limited. We characterized the anatomical, seasonal and developmental epidemiology of acute pediatric fracture admissions in a pooled hospital-based cohort. Methods: In this retrospective study, children (< 18 years) with a confirmed acute fracture admitted to three Chinese hospitals between October 2020 and November 2024 were identified from pooled electronic medical records by text mining of structured diagnoses; pathological and old fractures were excluded. Descriptive and regression analyses were patient-level (earliest acute admission per child); seasonality and secular trend in the fracture-admission rate, with all pediatric admissions as the denominator, were modeled at the episode level with cosinor negative-binomial regression. Multivariable logistic regression examined factors associated with multiple-region fracture and with lower-only (vs. upper-only) limb fracture, including the age-by-sex interaction and sex-specific site predilection. Results: Among 251,385 pediatric admissions, 7,142 (2.84%) were acute-fracture admissions in 6,206 children (70.9% male; male: female 2.44:1; mean age 9.7 ± 4.8 years). The upper limb was most often involved (48.9%), led by the humerus (21.6%) and forearm (15.6%); supracondylar humeral fractures comprised 9.7%. The fracture-admission rate showed autumn-peaking seasonality (negative-binomial peak-to-trough ratio 1.31, 95% CI 1.14–1.55; peak in mid-October; p = 0.007) and a declining fracture-admission rate relative to all pediatric admissions (rate ratio 0.82 per year, 95% CI 0.78–0.86). Lower-only (vs. upper-only) limb fracture rose with age (12–17 vs. 6–11 years: OR 2.30, 95% CI 2.00–2.65) and was less frequent in boys (OR 0.58, 95% CI 0.50–0.66); the sex effect varied with age (interaction p < 0.001). Site predilection was sex-specific (boys: forearm, hand; girls: ankle, foot, spine, pelvis, patella, tibia/fibula). Estimates were robust to the unit of analysis, limb coding, data-source stratum and denominator composition. Conclusions: In this hospital-based cohort, acute pediatric fracture admissions showed male and upper-limb predominance, an autumn peak, an age-dependent shift toward lower-limb injury, sex-specific site predilection and a declining share of pediatric admissions over time. Because injury mechanism, activity and a defined source population were unavailable, these patterns describe the hospital admission burden and generate hypotheses for age-, sex- and season-informed prevention research and trauma-service planning rather than establishing specific interventions.
- 10 October 2026
- Published by Springer Nature
- p. 73-92
AI Industry Development and Governance
- 10 October 2026
- Published by Springer Nature
- p. 1-26
AI Research and Development
- Open Access
- 25 August 2026
- Published by Elsevier in SSRN Electronic Journal
Sunshine-Induced Sentiment and Consumer Behaviors: New Evidence from Transaction-level Payments
This research investigates the casual impact of sunshine-induced sentiment on consumer behaviors by leveraging a novel high-frequency dataset of daily city-level weather conditions and individual transaction-level expenditures across over 100 categories in China. We document a sunshine effect: a one okta decrease in the same-day local sky cover is associated with a 0.09% increase in the daily total spending, which aggregates to an additional USD 12 billion in national annual expenditure after controlling for other weather conditions. This effect can be accounted for by an over-optimism mechanism. This sunshine effect is driven by non-discretionary rather than discretionary spending, suggesting that improved mood from sunshine leads consumers to spend more on necessities instead of impulsive purchases of non-essentials. Additionally, the effect is more pronounced among female consumers, those with lower wealth scores, and those with lower education levels. Finally, we also observe a significant but smaller positive effect on daily consumption frequency.
- 25 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSunshine-Induced Sentiment and Consumer Behaviors: New Evidence from Transaction-level Payments
This research investigates the casual impact of sunshine-induced sentiment on consumer behaviors by leveraging a novel high-frequency dataset of daily city-level weather conditions and individual transaction-level expenditures across over 100 categories in China. We document a sunshine effect: a one okta decrease in the same-day local sky cover is associated with a 0.09% increase in the daily total spending, which aggregates to an additional USD 12 billion in national annual expenditure after controlling for other weather conditions. This effect can be accounted for by an over-optimism mechanism. This sunshine effect is driven by non-discretionary rather than discretionary spending, suggesting that improved mood from sunshine leads consumers to spend more on necessities instead of impulsive purchases of non-essentials. Additionally, the effect is more pronounced among female consumers, those with lower wealth scores, and those with lower education levels. Finally, we also observe a significant but smaller positive effect on daily consumption frequency.
- Open Access
- 12 August 2026
- Published by MDPI AG in Preprints.org
Trustworthy Legal Reasoning in 2026: A Mid-Year Review of Research, Products, Events, and Governance
Legal artificial intelligence is moving beyond answer generation into consequential workflows spanning legal research, drafting, compliance, litigation support, public legal services, and regulated legal practice. This transition exposes the limits of conventional evaluation: a plausible answer may still rely on inappropriate authority, outdated law, an inapplicable jurisdiction, or an unreviewable process. We conduct a structured mid-year review of developments in trustworthy legal reasoning made public between 1 January and 30 June 2026. The study integrates four coded datasets: 141 topic-relevant publications meeting the review’s quality criteria, 99 product launch or major-update records, 53 curated public events, and 69 policy or regulatory records. In the publication sample, legal retrieval or retrieval-augmented generation appeared in 109 papers (77.3%), benchmark and evaluation research in 67 (47.5%), and legal agents or simulation in 30 (21.3%). Retrieval quality was addressed in 107 papers (75.9%), whereas temporal validity appeared in only 18 (12.8%) and uncertainty or refusal in nine (6.4%). Product activity was also geographically concentrated: developers headquartered in the United States and the United Kingdom accounted for 68.7% of observed records, while major updates outnumbered new launches. Across the four datasets, the field is moving toward workflow level, retrieval-grounded, and agentic legal AI, but temporal and jurisdictional validity, actionable uncertainty, reproducible oversight, and contestability remain unevenly operationalised. We argue that trustworthy legal reasoning is not a property of a model alone, but an institutional achievement requiring authoritative sources, inspectable processes, accountable human roles, and effective routes for challenge and correction.
- 12 August 2026
- Published by MDPI AG in Preprints.org
Open AccessTrustworthy Legal Reasoning in 2026: A Mid-Year Review of Research, Products, Events, and Governance
Legal artificial intelligence is moving beyond answer generation into consequential workflows spanning legal research, drafting, compliance, litigation support, public legal services, and regulated legal practice. This transition exposes the limits of conventional evaluation: a plausible answer may still rely on inappropriate authority, outdated law, an inapplicable jurisdiction, or an unreviewable process. We conduct a structured mid-year review of developments in trustworthy legal reasoning made public between 1 January and 30 June 2026. The study integrates four coded datasets: 141 topic-relevant publications meeting the review’s quality criteria, 99 product launch or major-update records, 53 curated public events, and 69 policy or regulatory records. In the publication sample, legal retrieval or retrieval-augmented generation appeared in 109 papers (77.3%), benchmark and evaluation research in 67 (47.5%), and legal agents or simulation in 30 (21.3%). Retrieval quality was addressed in 107 papers (75.9%), whereas temporal validity appeared in only 18 (12.8%) and uncertainty or refusal in nine (6.4%). Product activity was also geographically concentrated: developers headquartered in the United States and the United Kingdom accounted for 68.7% of observed records, while major updates outnumbered new launches. Across the four datasets, the field is moving toward workflow level, retrieval-grounded, and agentic legal AI, but temporal and jurisdictional validity, actionable uncertainty, reproducible oversight, and contestability remain unevenly operationalised. We argue that trustworthy legal reasoning is not a property of a model alone, but an institutional achievement requiring authoritative sources, inspectable processes, accountable human roles, and effective routes for challenge and correction.
- Open Access
- 10 August 2026
- Published by Elsevier in SSRN Electronic Journal
How Value Alignment Reorders Legal Reasons in Hybrid Legal Orders: Insights from Hong Kong Model Development
When "legal value alignment" is invoked in relation to generative AI, it is not a neutral promise of safer or more accurate outputs. It is a constitutional move in miniature. It quietly decides what will be treated as law, which values are permitted to matter, what forms of reasoning will be recognized as properly legal, and which sources will be allowed to speak with authority. Drawing on alignment insights from the Hong Kong Generative AI Research and Development Center and the governance practices that surround local model development and deployment, this Article argues that alignment in hybrid legal orders systematically produces a legitimacy illusion. A system can be tuned to sound compliant, coherent, and recognizably legal, while remaining unable to do what legitimacy requires in such settings: resolve conflicts of value and rank authority through institutionally authorized and publicly contestable justification. The unresolved remainder is not eliminated. It is relocated into technical choices that present themselves as mere implementation, including data curation, evaluation thresholds, safety boundaries, and refusal rules. Hong Kong brings this relocation into sharp relief. Common-law reason-giving and precedent preserve uncertainty as a condition of contestability, while national security governance can stabilize decisive premises upstream, compelling alignment to choose among rival interpretive grammars. The Article identifies four structural failure modes, categorical, logical, institutional, and epistemic, and explains how sensitivity labels and imported security taxonomies such as TC260-003 prefigure what may be said before legal reasoning begins. It concludes by theorizing four mechanisms that sustain the legitimacy illusion: contestation converted into defaults, authority laundered through model outputs, compliance aesthetics mistaken for legal justification, and polarized reception.
- 10 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessHow Value Alignment Reorders Legal Reasons in Hybrid Legal Orders: Insights from Hong Kong Model Development
When "legal value alignment" is invoked in relation to generative AI, it is not a neutral promise of safer or more accurate outputs. It is a constitutional move in miniature. It quietly decides what will be treated as law, which values are permitted to matter, what forms of reasoning will be recognized as properly legal, and which sources will be allowed to speak with authority. Drawing on alignment insights from the Hong Kong Generative AI Research and Development Center and the governance practices that surround local model development and deployment, this Article argues that alignment in hybrid legal orders systematically produces a legitimacy illusion. A system can be tuned to sound compliant, coherent, and recognizably legal, while remaining unable to do what legitimacy requires in such settings: resolve conflicts of value and rank authority through institutionally authorized and publicly contestable justification. The unresolved remainder is not eliminated. It is relocated into technical choices that present themselves as mere implementation, including data curation, evaluation thresholds, safety boundaries, and refusal rules. Hong Kong brings this relocation into sharp relief. Common-law reason-giving and precedent preserve uncertainty as a condition of contestability, while national security governance can stabilize decisive premises upstream, compelling alignment to choose among rival interpretive grammars. The Article identifies four structural failure modes, categorical, logical, institutional, and epistemic, and explains how sensitivity labels and imported security taxonomies such as TC260-003 prefigure what may be said before legal reasoning begins. It concludes by theorizing four mechanisms that sustain the legitimacy illusion: contestation converted into defaults, authority laundered through model outputs, compliance aesthetics mistaken for legal justification, and polarized reception.