Selected work

Representative Papers

Does ChatGPT score research quality differently by gender?

Aug 10, 2026

This study presents the first large-scale empirical investigation into whether large language models (LLMs), specifically ChatGPT, exhibit gender-based scoring bias when evaluating research quality under anonymized conditions. Leveraging 89,744 journal articles submitted to the UK’s Research Excellence Framework (REF) 2021, the authors removed author identifiers and obtained ChatGPT-assigned scores, which were then compared against official REF ratings and textual complexity metrics. Results indicate that papers with male first authors received slightly higher ChatGPT scores in most disciplines—particularly in health, science, and engineering—and this gender gap was more pronounced than in the official REF assessments. No significant disparity emerged in single-authored humanities and social sciences research. The findings suggest that the observed bias likely stems from indirect factors such as research field or topic rather than writing style, underscoring the need for caution regarding latent biases when deploying LLMs in research evaluation.

0 citationsRead paper

Augmenting Intelligence: A Hybrid Framework for Scalable and Stable Explanations

Dec 22, 2025

XAI faces a fundamental trade-off between scalability and stability: post-hoc methods (e.g., LIME, SHAP) scale well but yield unstable explanations, while supervised frameworks (e.g., TED) offer stability at the cost of heavy reliance on manual annotations. To resolve this, we propose LRR-TED—a hybrid framework integrating Generalized Linear Rule Models (GLRMs) with TED. It automatically discovers domain-agnostic “retention patterns” via rule learning and introduces two novel principles—“discovery asymmetry” and the “lost Anna Karenina principle”—enabling experts to annotate only critical churn-triggering rules, thereby shifting their role from rule authors to anomaly responders. Leveraging Pareto-optimal rule selection and dual-track modeling (safety net vs. risk trap), LRR-TED achieves 94.00% accuracy in customer churn prediction—outperforming eight handcrafted rule baselines while reducing annotation effort by 50%.

0 citationsRead paper

Lightweight Hopfield Neural Networks for Bioacoustic Detection and Call Monitoring of Captive Primates

Nov 04, 2025

Passive acoustic monitoring faces challenges of data processing latency, high annotation demands, and excessive computational costs associated with conventional CNN-based models. Method: This study proposes a lightweight, interpretable, and rapidly trainable Hopfield Neural Network (HNN) framework—the first application of HNNs to primate vocalization classification—specifically targeting social calls of captive black-and-white ruffed lemurs. The model integrates bat-inspired echolocation-motivated signal preprocessing and motion-correlated feature enhancement, enabling effective learning from minimal labeled samples. Training and real-time inference are achieved in milliseconds on standard laptop hardware. Contribution/Results: The framework achieves an overall accuracy of 0.94, a throughput of 340 frames per second, and processes over 5.5 hours of audio per minute. It drastically shortens the “data-to-decision” cycle, establishing a novel low-resource paradigm for bioacoustic monitoring in both field and captive settings.

0 citationsRead paper

A Fuzzy-Enhanced Explainable AI Framework for Flight Continuous Descent Operations Classification

Aug 20, 2025

Prior research lacks a systematic investigation of factors affecting Continuous Descent Operations (CDO) performance, and existing trajectory optimization methods suffer from limited interpretability. Method: This paper proposes FEXAI, an explainable AI framework integrating fuzzy logic, machine learning, and SHAP-based explanation techniques. Leveraging ADS-B data, we construct a 29-dimensional dataset comprising operational and meteorological features to enable CDO classification modeling and feature attribution analysis. Contribution/Results: FEXAI jointly enhances predictive accuracy and model transparency: all models achieve >90% classification accuracy; SHAP analysis identifies descent rate, number of descent segments, and heading change as the top three influential features; and interpretable, human-readable fuzzy rules are automatically generated to support real-time operational decision-making and safety validation. To our knowledge, this is the first work to deliver a high-accuracy, high-transparency, and production-deployable AI decision-support system for CDO in aviation.

0 citationsRead paper

First-of-its-kind AI model for bioacoustic detection using a lightweight associative memory Hopfield neural network

Jul 14, 2025

Bioacoustic monitoring faces three interrelated challenges: massive data volumes, high computational demands, and substantial environmental costs—including energy consumption and carbon footprint. To address these, we propose the first lightweight, transparent Hopfield associative memory model tailored for bioacoustic identification. Unlike conventional deep learning approaches, it requires no large-scale labeled datasets; instead, it achieves training using only a single representative exemplar—completing in 3 ms with 144.09 MB memory footprint. The model integrates signal preprocessing and similarity-matching mechanisms, enabling efficient deployment on edge devices. Evaluated on 10,384 bat recordings, it achieves end-to-end processing in 5.4 seconds with 86% accuracy, and its classifications fully align with expert annotations. This work pioneers the application of interpretable Hopfield networks in bioacoustics, uniquely balancing high accuracy, minimal resource consumption, and ecological sustainability.

0 citationsRead paper

Recent publications

Latest Papers

Does ChatGPT score research quality differently by gender?

Aug 10, 2026

This study presents the first large-scale empirical investigation into whether large language models (LLMs), specifically ChatGPT, exhibit gender-based scoring bias when evaluating research quality under anonymized conditions. Leveraging 89,744 journal articles submitted to the UK’s Research Excellence Framework (REF) 2021, the authors removed author identifiers and obtained ChatGPT-assigned scores, which were then compared against official REF ratings and textual complexity metrics. Results indicate that papers with male first authors received slightly higher ChatGPT scores in most disciplines—particularly in health, science, and engineering—and this gender gap was more pronounced than in the official REF assessments. No significant disparity emerged in single-authored humanities and social sciences research. The findings suggest that the observed bias likely stems from indirect factors such as research field or topic rather than writing style, underscoring the need for caution regarding latent biases when deploying LLMs in research evaluation.

0 citationsRead paper

Augmenting Intelligence: A Hybrid Framework for Scalable and Stable Explanations

Dec 22, 2025

XAI faces a fundamental trade-off between scalability and stability: post-hoc methods (e.g., LIME, SHAP) scale well but yield unstable explanations, while supervised frameworks (e.g., TED) offer stability at the cost of heavy reliance on manual annotations. To resolve this, we propose LRR-TED—a hybrid framework integrating Generalized Linear Rule Models (GLRMs) with TED. It automatically discovers domain-agnostic “retention patterns” via rule learning and introduces two novel principles—“discovery asymmetry” and the “lost Anna Karenina principle”—enabling experts to annotate only critical churn-triggering rules, thereby shifting their role from rule authors to anomaly responders. Leveraging Pareto-optimal rule selection and dual-track modeling (safety net vs. risk trap), LRR-TED achieves 94.00% accuracy in customer churn prediction—outperforming eight handcrafted rule baselines while reducing annotation effort by 50%.

0 citationsRead paper

Lightweight Hopfield Neural Networks for Bioacoustic Detection and Call Monitoring of Captive Primates

Nov 04, 2025

Passive acoustic monitoring faces challenges of data processing latency, high annotation demands, and excessive computational costs associated with conventional CNN-based models. Method: This study proposes a lightweight, interpretable, and rapidly trainable Hopfield Neural Network (HNN) framework—the first application of HNNs to primate vocalization classification—specifically targeting social calls of captive black-and-white ruffed lemurs. The model integrates bat-inspired echolocation-motivated signal preprocessing and motion-correlated feature enhancement, enabling effective learning from minimal labeled samples. Training and real-time inference are achieved in milliseconds on standard laptop hardware. Contribution/Results: The framework achieves an overall accuracy of 0.94, a throughput of 340 frames per second, and processes over 5.5 hours of audio per minute. It drastically shortens the “data-to-decision” cycle, establishing a novel low-resource paradigm for bioacoustic monitoring in both field and captive settings.

0 citationsRead paper

A Fuzzy-Enhanced Explainable AI Framework for Flight Continuous Descent Operations Classification

Aug 20, 2025

Prior research lacks a systematic investigation of factors affecting Continuous Descent Operations (CDO) performance, and existing trajectory optimization methods suffer from limited interpretability. Method: This paper proposes FEXAI, an explainable AI framework integrating fuzzy logic, machine learning, and SHAP-based explanation techniques. Leveraging ADS-B data, we construct a 29-dimensional dataset comprising operational and meteorological features to enable CDO classification modeling and feature attribution analysis. Contribution/Results: FEXAI jointly enhances predictive accuracy and model transparency: all models achieve >90% classification accuracy; SHAP analysis identifies descent rate, number of descent segments, and heading change as the top three influential features; and interpretable, human-readable fuzzy rules are automatically generated to support real-time operational decision-making and safety validation. To our knowledge, this is the first work to deliver a high-accuracy, high-transparency, and production-deployable AI decision-support system for CDO in aviation.

0 citationsRead paper

First-of-its-kind AI model for bioacoustic detection using a lightweight associative memory Hopfield neural network

Jul 14, 2025

Bioacoustic monitoring faces three interrelated challenges: massive data volumes, high computational demands, and substantial environmental costs—including energy consumption and carbon footprint. To address these, we propose the first lightweight, transparent Hopfield associative memory model tailored for bioacoustic identification. Unlike conventional deep learning approaches, it requires no large-scale labeled datasets; instead, it achieves training using only a single representative exemplar—completing in 3 ms with 144.09 MB memory footprint. The model integrates signal preprocessing and similarity-matching mechanisms, enabling efficient deployment on edge devices. Evaluated on 10,384 bat recordings, it achieves end-to-end processing in 5.4 seconds with 86% accuracy, and its classifications fully align with expert annotations. This work pioneers the application of interpretable Hopfield networks in bioacoustics, uniquely balancing high accuracy, minimal resource consumption, and ecological sustainability.

0 citationsRead paper