A spatio-temporal block aggregation model for latent log Gaussian outcomes: application on modelling wastewater virus concentration in Wales
本文提出了一种时空块聚合模型，用于处理废水病毒浓度的潜在对数高斯结果，并通过贝叶斯框架和INLA方法进行推断，应用于威尔士47个集水区的SARS-CoV-2监测。
0 citationsRead paper
Selected work
Representative Papers
Judges matter more than papers in post-publication research assessment
This study addresses the substantial impact of reviewer subjectivity on the reliability of scholarly evaluations. Leveraging 239,521 peer review records from the H1 Connect platform, the authors employ multilevel linear modeling and variance decomposition to systematically quantify the dominant contribution of reviewer-related variability for the first time. The results reveal that reviewer-level effects account for 61% of the total variance in scores—far exceeding the combined influence of manuscript and journal factors (7%) and demographic or institutional biases such as author gender or affiliation (<1%). These findings demonstrate that “reviewer noise” constitutes the primary source of evaluation bias, prompting the authors to advocate for the implementation of “noise audits” in high-stakes academic assessments to enhance fairness and scientific rigor.
0 citationsRead paper
Dynamic Bidirectional Pattern Memory: A Production-Scale Empirical Characterisation of Inference-Time Gating in Clinical NLP
This work addresses the inefficiency of repeatedly verifying redundant or invalid information in clinical NLP generation–verification pipelines by proposing a lightweight dynamic memory mechanism that learns gating strategies during inference to filter spurious outputs. The study finds that directly learning filtering rules from sparse rejection signals provided by verifiers yields limited performance; instead, integrating clinical ontologies with evidence detection aligned to verification logic enables effective gating. The system employs a dual-model architecture comprising a Llama-3.3 70B generator and an MMed-Llama-3.1 70B verifier, along with a flag-but-not-delete policy for handling suspicious outputs to preserve clinical auditability. Experiments on 5,000 patient records demonstrate that the ontology-driven filter identifies 49,734 violation relations, while the question-answering filter increases the verifier’s rejection probability by 1.84×.
0 citationsRead paper
How much of an LLM-generated clinical corpus is actually new? A production-scale measurement of content redundancy for provenance classification
This study addresses a common misconception in clinical machine learning that equates the sheer scale of large language model (LLM)-generated text with its informational content, often overlooking pervasive redundancy. The authors propose a source-based redundancy decomposition method to classify each token within 2.51 billion tokens extracted via multi-agent LLM processing from 167,034 patient narratives. At production scale, they reveal that only 10.9% of tokens represent unique content, while 79.4% are redundant—indicating that raw token counts overestimate true information by approximately ninefold. The work further distinguishes two distinct redundancy mechanisms and demonstrates that redundancy stems from pipeline design rather than inherent LLM properties. Through lossless compression analysis and controlled fine-tuning experiments, the study shows that deduplication significantly improves clinical encoder performance on external disease recognition tasks under identical token budgets.
0 citationsRead paper
Embedding Empathy into Visual Analytics: A Framework for Person-Centred Dementia Care
Current digital dementia care tools over-rely on quantitative metrics while neglecting empathic dimensions, impeding caregivers’ ability to perceive individual patient needs and emotional cues. To address this, we propose the first design framework that systematically embeds empathy into visual analytics—integrating empathy mapping, patient-centered care principles, and user-centered design methods to jointly optimize experiences for clinical staff (primary users) and patients (secondary users). Through iterative visual design research, usability testing, and expert evaluation by healthcare professionals, our prototype demonstrably strengthens clinician–patient empathic connection. Results validate its feasibility in enhancing human-centeredness, ethical alignment, and clinical applicability of dementia care. Our core contribution is the novel conceptualization of empathy as a foundational mechanism in visualization design—advancing health information visualization toward greater humanistic sensitivity and relational intelligence.
0 citationsRead paper
Recent publications
Latest Papers
A spatio-temporal block aggregation model for latent log Gaussian outcomes: application on modelling wastewater virus concentration in Wales
本文提出了一种时空块聚合模型，用于处理废水病毒浓度的潜在对数高斯结果，并通过贝叶斯框架和INLA方法进行推断，应用于威尔士47个集水区的SARS-CoV-2监测。
0 citationsRead paper
Judges matter more than papers in post-publication research assessment
This study addresses the substantial impact of reviewer subjectivity on the reliability of scholarly evaluations. Leveraging 239,521 peer review records from the H1 Connect platform, the authors employ multilevel linear modeling and variance decomposition to systematically quantify the dominant contribution of reviewer-related variability for the first time. The results reveal that reviewer-level effects account for 61% of the total variance in scores—far exceeding the combined influence of manuscript and journal factors (7%) and demographic or institutional biases such as author gender or affiliation (<1%). These findings demonstrate that “reviewer noise” constitutes the primary source of evaluation bias, prompting the authors to advocate for the implementation of “noise audits” in high-stakes academic assessments to enhance fairness and scientific rigor.
0 citationsRead paper
Dynamic Bidirectional Pattern Memory: A Production-Scale Empirical Characterisation of Inference-Time Gating in Clinical NLP
This work addresses the inefficiency of repeatedly verifying redundant or invalid information in clinical NLP generation–verification pipelines by proposing a lightweight dynamic memory mechanism that learns gating strategies during inference to filter spurious outputs. The study finds that directly learning filtering rules from sparse rejection signals provided by verifiers yields limited performance; instead, integrating clinical ontologies with evidence detection aligned to verification logic enables effective gating. The system employs a dual-model architecture comprising a Llama-3.3 70B generator and an MMed-Llama-3.1 70B verifier, along with a flag-but-not-delete policy for handling suspicious outputs to preserve clinical auditability. Experiments on 5,000 patient records demonstrate that the ontology-driven filter identifies 49,734 violation relations, while the question-answering filter increases the verifier’s rejection probability by 1.84×.
0 citationsRead paper
How much of an LLM-generated clinical corpus is actually new? A production-scale measurement of content redundancy for provenance classification
This study addresses a common misconception in clinical machine learning that equates the sheer scale of large language model (LLM)-generated text with its informational content, often overlooking pervasive redundancy. The authors propose a source-based redundancy decomposition method to classify each token within 2.51 billion tokens extracted via multi-agent LLM processing from 167,034 patient narratives. At production scale, they reveal that only 10.9% of tokens represent unique content, while 79.4% are redundant—indicating that raw token counts overestimate true information by approximately ninefold. The work further distinguishes two distinct redundancy mechanisms and demonstrates that redundancy stems from pipeline design rather than inherent LLM properties. Through lossless compression analysis and controlled fine-tuning experiments, the study shows that deduplication significantly improves clinical encoder performance on external disease recognition tasks under identical token budgets.
0 citationsRead paper
Embedding Empathy into Visual Analytics: A Framework for Person-Centred Dementia Care
Current digital dementia care tools over-rely on quantitative metrics while neglecting empathic dimensions, impeding caregivers’ ability to perceive individual patient needs and emotional cues. To address this, we propose the first design framework that systematically embeds empathy into visual analytics—integrating empathy mapping, patient-centered care principles, and user-centered design methods to jointly optimize experiences for clinical staff (primary users) and patients (secondary users). Through iterative visual design research, usability testing, and expert evaluation by healthcare professionals, our prototype demonstrably strengthens clinician–patient empathic connection. Results validate its feasibility in enhancing human-centeredness, ethical alignment, and clinical applicability of dementia care. Our core contribution is the novel conceptualization of empathy as a foundational mechanism in visualization design—advancing health information visualization toward greater humanistic sensitivity and relational intelligence.
0 citationsRead paper