Selected work

Representative Papers

Dynamical behaviors of a stochastic SIS epidemic model with mean-reverting inhomogeneous geometric brownian motion

Mar 24, 2026

The main purpose of this paper is to study the Dynamical behaviors of a stochastic SIS epidemic model using mean-reverting inhomogeneous geometric brownian motion process. First we demonstrate the existence of a global-in-time solution and establish that is unique and remains positive. Then we derive a sufficient condition for exponential extinction of infectious diseases and we show that our extinction threshold in the stochastic case coincides with that of the deterministic case. Finaly, we define an appropriate theoretical framework to guarantee the existence of an ergodic stationary distribution.

0 citationsRead paper

Efficient Topic Extraction via Graph-Based Labeling: A Lightweight Alternative to Deep Models

Nov 06, 2025

To address the high computational cost and poor interpretability of deep models in topic modeling, this paper proposes a lightweight graph-enhanced topic labeling method. Unlike LLM-dependent approaches, it constructs a semantic word graph over pre-trained word embeddings, leverages graph propagation to expand topic vocabularies and uncover structured semantic relationships among words, and generates concise, semantically coherent topic labels. Experiments on two standard benchmarks demonstrate that our method significantly outperforms traditional baselines (e.g., LDA+KeyBERT) in label quality—measured by BERTScore and cosine similarity—and matches ChatGPT-3.5’s performance, while achieving 3–5× faster inference and reducing memory consumption by ~70%. Our core contribution is the first integration of graph-structured modeling into topic labeling, enabling efficient, low-resource, and highly interpretable automatic label generation.

0 citationsRead paper

Improving Diagnostic Performance on Small and Imbalanced Datasets Using Class-Based Input Image Composition

Nov 05, 2025

To address misdiagnosis in medical deep learning caused by few-shot learning, severe class imbalance, and low-quality imaging, this paper proposes a Class-Based Image Composition (CBIC) method. CBIC synthesizes multiple OCT images of the same class into a 3×1 composite input (CoImg), enabling construction of a balanced dataset—Co-OCTDL—without modifying model architecture or hyperparameters. Applied directly to VGG16, CBIC enhances intra-class diversity and information density, thereby improving discrimination of subtle pathological patterns. On the OCTDL dataset, it achieves 99.6% accuracy, an F1-score of 0.995, and an AUC of 0.9996, significantly reducing misclassification rates. The core contribution is an unsupervised, lightweight image fusion strategy that mitigates data weakness in small-sample, imbalanced medical imaging scenarios, yielding a reproducible, plug-and-play diagnostic paradigm.

0 citationsRead paper

Comprehensive Analysis of VQC for Financial Fraud Detection: A Comparative Study of Quantum Encoding Techniques and Architectural Optimizations

Sep 26, 2025

Variational quantum classifiers (VQCs) lack systematic performance analysis for high-stakes financial fraud detection, particularly regarding the impact of quantum circuit design choices on model expressivity and trainability. Method: We conduct controlled experiments varying quantum encoding schemes (e.g., ZZ-encoding), entanglement topologies (ring, linear, fully connected), circuit depth, and optimization strategies. We further introduce a novel quantum circuit visualization framework to analyze how entanglement structure governs the trade-off between model expressivity and trainability. Contribution/Results: Ring topology achieves the optimal balance between accuracy and training stability, yielding 93.3% classification accuracy; the overall best configuration reaches 94.3%, significantly outperforming classical baselines. This work provides empirically grounded deployment guidelines for VQCs in financial fraud detection and—crucially—establishes entanglement topology as an explicit, tunable performance determinant for the first time, advancing interpretable quantum machine learning in financial security applications.

0 citationsRead paper

PlantBert: An Open Source Language Model for Plant Science

Jun 10, 2025

The plant sciences lack domain-specific language models capable of efficiently extracting structured knowledge from stress-response literature. Method: We introduce the first open-source, domain-adapted language model for plant stress responses, specifically targeting molecular, physiological, and agronomic responses of chickpea to abiotic and biotic stresses. Our approach deeply adapts the DeBERTa architecture to the plant science domain and integrates rule-enhanced post-processing with Crop Ontology–driven entity normalization to achieve high-precision semantic relation modeling under low-resource conditions. The model is trained on expert-annotated corpora aligned with a hierarchical ontology-based annotation schema. Contribution/Results: It significantly outperforms general-purpose models across multiple biomedical entity recognition tasks and demonstrates strong cross-task generalization. The code and pretrained models are publicly released to advance intelligent knowledge mining in plant genomics, phenomics, and agronomy.

0 citationsRead paper

Recent publications

Latest Papers

Dynamical behaviors of a stochastic SIS epidemic model with mean-reverting inhomogeneous geometric brownian motion

Mar 24, 2026

The main purpose of this paper is to study the Dynamical behaviors of a stochastic SIS epidemic model using mean-reverting inhomogeneous geometric brownian motion process. First we demonstrate the existence of a global-in-time solution and establish that is unique and remains positive. Then we derive a sufficient condition for exponential extinction of infectious diseases and we show that our extinction threshold in the stochastic case coincides with that of the deterministic case. Finaly, we define an appropriate theoretical framework to guarantee the existence of an ergodic stationary distribution.

0 citationsRead paper

Efficient Topic Extraction via Graph-Based Labeling: A Lightweight Alternative to Deep Models

Nov 06, 2025

To address the high computational cost and poor interpretability of deep models in topic modeling, this paper proposes a lightweight graph-enhanced topic labeling method. Unlike LLM-dependent approaches, it constructs a semantic word graph over pre-trained word embeddings, leverages graph propagation to expand topic vocabularies and uncover structured semantic relationships among words, and generates concise, semantically coherent topic labels. Experiments on two standard benchmarks demonstrate that our method significantly outperforms traditional baselines (e.g., LDA+KeyBERT) in label quality—measured by BERTScore and cosine similarity—and matches ChatGPT-3.5’s performance, while achieving 3–5× faster inference and reducing memory consumption by ~70%. Our core contribution is the first integration of graph-structured modeling into topic labeling, enabling efficient, low-resource, and highly interpretable automatic label generation.

0 citationsRead paper

Improving Diagnostic Performance on Small and Imbalanced Datasets Using Class-Based Input Image Composition

Nov 05, 2025

To address misdiagnosis in medical deep learning caused by few-shot learning, severe class imbalance, and low-quality imaging, this paper proposes a Class-Based Image Composition (CBIC) method. CBIC synthesizes multiple OCT images of the same class into a 3×1 composite input (CoImg), enabling construction of a balanced dataset—Co-OCTDL—without modifying model architecture or hyperparameters. Applied directly to VGG16, CBIC enhances intra-class diversity and information density, thereby improving discrimination of subtle pathological patterns. On the OCTDL dataset, it achieves 99.6% accuracy, an F1-score of 0.995, and an AUC of 0.9996, significantly reducing misclassification rates. The core contribution is an unsupervised, lightweight image fusion strategy that mitigates data weakness in small-sample, imbalanced medical imaging scenarios, yielding a reproducible, plug-and-play diagnostic paradigm.

0 citationsRead paper

Comprehensive Analysis of VQC for Financial Fraud Detection: A Comparative Study of Quantum Encoding Techniques and Architectural Optimizations

Sep 26, 2025

Variational quantum classifiers (VQCs) lack systematic performance analysis for high-stakes financial fraud detection, particularly regarding the impact of quantum circuit design choices on model expressivity and trainability. Method: We conduct controlled experiments varying quantum encoding schemes (e.g., ZZ-encoding), entanglement topologies (ring, linear, fully connected), circuit depth, and optimization strategies. We further introduce a novel quantum circuit visualization framework to analyze how entanglement structure governs the trade-off between model expressivity and trainability. Contribution/Results: Ring topology achieves the optimal balance between accuracy and training stability, yielding 93.3% classification accuracy; the overall best configuration reaches 94.3%, significantly outperforming classical baselines. This work provides empirically grounded deployment guidelines for VQCs in financial fraud detection and—crucially—establishes entanglement topology as an explicit, tunable performance determinant for the first time, advancing interpretable quantum machine learning in financial security applications.

0 citationsRead paper

PlantBert: An Open Source Language Model for Plant Science

Jun 10, 2025

The plant sciences lack domain-specific language models capable of efficiently extracting structured knowledge from stress-response literature. Method: We introduce the first open-source, domain-adapted language model for plant stress responses, specifically targeting molecular, physiological, and agronomic responses of chickpea to abiotic and biotic stresses. Our approach deeply adapts the DeBERTa architecture to the plant science domain and integrates rule-enhanced post-processing with Crop Ontology–driven entity normalization to achieve high-precision semantic relation modeling under low-resource conditions. The model is trained on expert-annotated corpora aligned with a hierarchical ontology-based annotation schema. Contribution/Results: It significantly outperforms general-purpose models across multiple biomedical entity recognition tasks and demonstrates strong cross-task generalization. The code and pretrained models are publicly released to advance intelligent knowledge mining in plant genomics, phenomics, and agronomy.

0 citationsRead paper