Selected work

Representative Papers

Clinical Knowledge Graph Construction and Evaluation with Multi-LLMs via Retrieval-Augmented Generation

Jan 05, 2026arXiv.org

This work addresses the limitations of existing approaches in constructing oncology knowledge graphs from unstructured clinical text, which often lack effective fact verification and semantic consistency. The authors propose an end-to-end KG-RAG framework that integrates multi-agent prompt engineering, retrieval-augmented generation, and ontology-aligned RDF/OWL semantic modeling to directly extract entities, attributes, and relations. To mitigate hallucination and enhance semantic fidelity, the method incorporates an entropy-based uncertainty scoring mechanism and a multi-LLM consensus strategy. Notably, it enables gold-standard-free, self-supervised continuous refinement. Evaluated on PDAC and BRCA patient cohorts, the resulting knowledge graphs demonstrate high clinical credibility, SPARQL compatibility, and significant improvements over baseline methods in precision, relevance, and ontological compliance.

1 citationsRead paper

Lightweight Transformer Models for On-Device Fault Detection: A Benchmark Study on Resource-Constrained Deployment

Jun 23, 2026

This study addresses the trade-offs among accuracy, latency, and model size when deploying fault detection models on resource-constrained devices, particularly under extreme class imbalance where existing approaches often underperform. The work presents the first systematic comparison of lightweight Transformers—including DistilBERT, TinyBERT, and MobileBERT—against traditional machine learning models such as XGBoost and SVM across three public datasets. Furthermore, it proposes an efficient deployment strategy combining INT8 dynamic quantization with two-stage adaptive inference. Experimental results show that TinyBERT-4L achieves an F1 score of 87.8% on the C-MAPSS dataset; post-quantization, model size is reduced by 25% with only a marginal F1 drop to 86.9%. The adaptive inference scheme attains an F1 of 87.6% at an average latency of 19.5 ms, though performance degrades significantly on highly imbalanced data.

0 citationsRead paper

LLM-Enhanced Log Anomaly Detection: A Comprehensive Benchmark of Large Language Models for Automated System Diagnostics

Apr 13, 2026

This study addresses the challenges posed by the heterogeneity and dynamic evolution of modern system logs, which hinder traditional anomaly detection approaches, and fills a critical gap in the systematic evaluation of large language models (LLMs) for this task. It presents the first comprehensive comparison among classical parser-plus-classifier pipelines, fine-tuned Transformers (e.g., BERT, RoBERTa), and prompt-driven LLMs (e.g., GPT-3.5, GPT-4, LLaMA-3) across four public log datasets. Results demonstrate that fine-tuned models achieve state-of-the-art F1 scores (0.96–0.99), while zero-shot and few-shot LLMs also exhibit strong performance (F1: 0.82–0.91), highlighting their practical advantage in label-scarce scenarios. The work further offers actionable insights into trade-offs among accuracy, inference latency, cost, and label dependency, with all code and configurations publicly released.

0 citationsRead paper

Selective Fine-Tuning of GPT Architectures for Parameter-Efficient Clinical Text Classification

Mar 14, 2026

This study addresses the challenges of clinical text classification, which include scarce annotated data, complex medical terminology, and the high computational cost of full-parameter fine-tuning of large language models. To overcome these limitations, the authors propose a parameter-efficient selective fine-tuning approach for GPT-2 that updates only the final Transformer block, layer normalization modules, and a lightweight classification head. This strategy preserves the model’s pretrained linguistic representations while reducing trainable parameters to less than 6% of the original. Evaluated on 50,000 radiology reports from the MIMIC-IV-Note dataset with CheXpert-derived labels, the method achieves approximately 91% accuracy—significantly outperforming both classifier-only training and full-model fine-tuning baselines—thus offering an effective balance between performance and computational efficiency.

0 citationsRead paper

Budget Forecasting and Integrated Strategic Planning for Leaders

Oct 29, 2025

This study addresses the persistent challenge faced by California Community Colleges (CCCs) in aligning financial planning with their Diversity, Equity, and Inclusion (DEI) mission amid ongoing state budget reforms. Using publicly available data from 1993 to 2023, we employ quantitative correlation analysis and statistical modeling to assess the predictive power of macroeconomic indicators on state education appropriations. Results indicate that GDP growth rate and the Consumer Price Index (CPI) exert statistically significant positive effects on state-level community college funding (p < 0.01). Building on these findings, we propose an institutional innovation: integrating a dynamic economic forecasting system into colleges’ strategic decision-making frameworks. This approach enhances budgetary resource allocation efficiency and strengthens sustainable support for historically underserved student populations. The study contributes empirically grounded, actionable policy and administrative design recommendations for advancing fiscal resilience and equity-centered resource stewardship in public higher education.

0 citationsRead paper

Recent publications

Latest Papers

Lightweight Transformer Models for On-Device Fault Detection: A Benchmark Study on Resource-Constrained Deployment

Jun 23, 2026

This study addresses the trade-offs among accuracy, latency, and model size when deploying fault detection models on resource-constrained devices, particularly under extreme class imbalance where existing approaches often underperform. The work presents the first systematic comparison of lightweight Transformers—including DistilBERT, TinyBERT, and MobileBERT—against traditional machine learning models such as XGBoost and SVM across three public datasets. Furthermore, it proposes an efficient deployment strategy combining INT8 dynamic quantization with two-stage adaptive inference. Experimental results show that TinyBERT-4L achieves an F1 score of 87.8% on the C-MAPSS dataset; post-quantization, model size is reduced by 25% with only a marginal F1 drop to 86.9%. The adaptive inference scheme attains an F1 of 87.6% at an average latency of 19.5 ms, though performance degrades significantly on highly imbalanced data.

0 citationsRead paper

LLM-Enhanced Log Anomaly Detection: A Comprehensive Benchmark of Large Language Models for Automated System Diagnostics

Apr 13, 2026

This study addresses the challenges posed by the heterogeneity and dynamic evolution of modern system logs, which hinder traditional anomaly detection approaches, and fills a critical gap in the systematic evaluation of large language models (LLMs) for this task. It presents the first comprehensive comparison among classical parser-plus-classifier pipelines, fine-tuned Transformers (e.g., BERT, RoBERTa), and prompt-driven LLMs (e.g., GPT-3.5, GPT-4, LLaMA-3) across four public log datasets. Results demonstrate that fine-tuned models achieve state-of-the-art F1 scores (0.96–0.99), while zero-shot and few-shot LLMs also exhibit strong performance (F1: 0.82–0.91), highlighting their practical advantage in label-scarce scenarios. The work further offers actionable insights into trade-offs among accuracy, inference latency, cost, and label dependency, with all code and configurations publicly released.

0 citationsRead paper

Selective Fine-Tuning of GPT Architectures for Parameter-Efficient Clinical Text Classification

Mar 14, 2026

This study addresses the challenges of clinical text classification, which include scarce annotated data, complex medical terminology, and the high computational cost of full-parameter fine-tuning of large language models. To overcome these limitations, the authors propose a parameter-efficient selective fine-tuning approach for GPT-2 that updates only the final Transformer block, layer normalization modules, and a lightweight classification head. This strategy preserves the model’s pretrained linguistic representations while reducing trainable parameters to less than 6% of the original. Evaluated on 50,000 radiology reports from the MIMIC-IV-Note dataset with CheXpert-derived labels, the method achieves approximately 91% accuracy—significantly outperforming both classifier-only training and full-model fine-tuning baselines—thus offering an effective balance between performance and computational efficiency.

0 citationsRead paper

Clinical Knowledge Graph Construction and Evaluation with Multi-LLMs via Retrieval-Augmented Generation

Jan 05, 2026arXiv.org

This work addresses the limitations of existing approaches in constructing oncology knowledge graphs from unstructured clinical text, which often lack effective fact verification and semantic consistency. The authors propose an end-to-end KG-RAG framework that integrates multi-agent prompt engineering, retrieval-augmented generation, and ontology-aligned RDF/OWL semantic modeling to directly extract entities, attributes, and relations. To mitigate hallucination and enhance semantic fidelity, the method incorporates an entropy-based uncertainty scoring mechanism and a multi-LLM consensus strategy. Notably, it enables gold-standard-free, self-supervised continuous refinement. Evaluated on PDAC and BRCA patient cohorts, the resulting knowledge graphs demonstrate high clinical credibility, SPARQL compatibility, and significant improvements over baseline methods in precision, relevance, and ontological compliance.

1 citationsRead paper

Budget Forecasting and Integrated Strategic Planning for Leaders

Oct 29, 2025

This study addresses the persistent challenge faced by California Community Colleges (CCCs) in aligning financial planning with their Diversity, Equity, and Inclusion (DEI) mission amid ongoing state budget reforms. Using publicly available data from 1993 to 2023, we employ quantitative correlation analysis and statistical modeling to assess the predictive power of macroeconomic indicators on state education appropriations. Results indicate that GDP growth rate and the Consumer Price Index (CPI) exert statistically significant positive effects on state-level community college funding (p < 0.01). Building on these findings, we propose an institutional innovation: integrating a dynamic economic forecasting system into colleges’ strategic decision-making frameworks. This approach enhances budgetary resource allocation efficiency and strengthens sustainable support for historically underserved student populations. The study contributes empirically grounded, actionable policy and administrative design recommendations for advancing fiscal resilience and equity-centered resource stewardship in public higher education.

0 citationsRead paper