Selected work

Representative Papers

Towards eco friendly cybersecurity: machine learning based anomaly detection with carbon and energy metrics

Nov 03, 2025International Journal of Applied Mathematics

This study addresses a critical gap in cybersecurity research by incorporating energy consumption and carbon emissions into the evaluation of AI-based anomaly detection systems, an aspect largely overlooked in existing literature. The authors propose the first green framework that integrates environmental impact metrics into network intrusion detection assessment, introducing an "Eco-Efficiency Index" to jointly quantify model performance and ecological cost. Leveraging Logistic Regression, Random Forest, SVM, Isolation Forest, and XGBoost, the framework employs CodeCarbon for carbon tracking and principal component analysis for energy-efficiency optimization. Experimental results demonstrate that the optimized Random Forest and lightweight Logistic Regression models achieve high detection accuracy while reducing energy consumption by over 40% compared to XGBoost, thereby substantiating the feasibility of environmentally sustainable cybersecurity solutions.

6 citationsRead paper

Rough Volatility Across Assets

Aug 17, 2026

This study addresses inconsistencies and estimation biases in cross-asset volatility roughness measurements by proposing a mean-reversion contamination correction formula and an additive noise adjustment framework, supported by a unified data infrastructure and a method applicability taxonomy. Empirical results confirm that realized volatility is universally rough across asset classes, with corrected Hurst exponents remaining significantly below the Brownian diffusion benchmark. Furthermore, the analysis reveals that implied roughness estimates for equity indices slightly exceed realized values, whereas such estimates fail for interest rates and foreign exchange. By effectively mitigating noise interference in rough volatility modeling, this work clarifies the systematic discrepancies between implied and realized roughness measures, providing a robust foundation for cross-asset volatility analysis.

0 citationsRead paper

Attention Without Grounding: Causal Evaluation of Visual Explanations in Medical VLMs

Jul 20, 2026

Current medical vision-language models (VLMs) produce attention heatmaps that lack causal validation, making it difficult to ascertain whether these maps genuinely reflect the critical image regions underlying model predictions. This work proposes the first multidimensional evaluation framework integrating clinical annotations with causal perturbations to systematically assess the faithfulness of VLM attention. The framework evaluates region overlap with radiologist-annotated areas, attribution quality within masked regions, and performance under 16×16 image patch occlusion. Results reveal that none of the evaluated VLMs simultaneously satisfy the dual criteria of effectively leveraging visual information and concentrating attention on clinically relevant regions. In contrast, all specialized chest X-ray (CXR) classifiers pass the assessment, exposing a fundamental deficiency in the explainability of existing medical VLMs.

0 citationsRead paper

An Explainable Agentic System for Detection of Conversational Scams with Summary-Based Memory

Jul 13, 2026

This study addresses the limitations of existing fraud detection systems, which typically analyze isolated messages and struggle to identify sophisticated scams that unfold over weeks or months through progressive trust-building dialogues. To tackle this challenge, the work proposes the first explainable agent architecture tailored for multi-category conversational fraud detection, integrating a summary-based memory mechanism with a dual-level (message- and conversation-level) detection framework that combines conversational semantic modeling and explainable AI techniques. The contributions include the release of ConScamBench-278, the first public multi-category benchmark dataset for conversational scams, achieving a 100% scam conversation detection rate on LoveFraud02 and 97.8% accuracy on ConScamBench-278. User studies further demonstrate that the system significantly enhances user trust and usability, attaining a usability score of 74.7.

0 citationsRead paper

DS-SAC: Density Search for Sample Consensus

Jul 04, 2026

This work addresses the overreliance on random sampling in robust geometric model estimation for computer vision by proposing a deterministic robust estimation framework. The method introduces, for the first time, deterministic dense-region search into the sample consensus paradigm, efficiently exploring high-consensus models through an initial-model-guided local forward-backward search combined with recursive partitioning based on signed residuals. This approach eliminates randomness while guaranteeing polynomial time complexity. Evaluated on homography, fundamental matrix, and essential matrix estimation tasks, the proposed method consistently outperforms mainstream approaches such as RANSAC and MAGSAC in terms of area under the cumulative error curve (AUC), median pose error, and computational speed.

0 citationsRead paper

Recent publications

Latest Papers

Rough Volatility Across Assets

Aug 17, 2026

This study addresses inconsistencies and estimation biases in cross-asset volatility roughness measurements by proposing a mean-reversion contamination correction formula and an additive noise adjustment framework, supported by a unified data infrastructure and a method applicability taxonomy. Empirical results confirm that realized volatility is universally rough across asset classes, with corrected Hurst exponents remaining significantly below the Brownian diffusion benchmark. Furthermore, the analysis reveals that implied roughness estimates for equity indices slightly exceed realized values, whereas such estimates fail for interest rates and foreign exchange. By effectively mitigating noise interference in rough volatility modeling, this work clarifies the systematic discrepancies between implied and realized roughness measures, providing a robust foundation for cross-asset volatility analysis.

0 citationsRead paper

Attention Without Grounding: Causal Evaluation of Visual Explanations in Medical VLMs

Jul 20, 2026

Current medical vision-language models (VLMs) produce attention heatmaps that lack causal validation, making it difficult to ascertain whether these maps genuinely reflect the critical image regions underlying model predictions. This work proposes the first multidimensional evaluation framework integrating clinical annotations with causal perturbations to systematically assess the faithfulness of VLM attention. The framework evaluates region overlap with radiologist-annotated areas, attribution quality within masked regions, and performance under 16×16 image patch occlusion. Results reveal that none of the evaluated VLMs simultaneously satisfy the dual criteria of effectively leveraging visual information and concentrating attention on clinically relevant regions. In contrast, all specialized chest X-ray (CXR) classifiers pass the assessment, exposing a fundamental deficiency in the explainability of existing medical VLMs.

0 citationsRead paper

An Explainable Agentic System for Detection of Conversational Scams with Summary-Based Memory

Jul 13, 2026

This study addresses the limitations of existing fraud detection systems, which typically analyze isolated messages and struggle to identify sophisticated scams that unfold over weeks or months through progressive trust-building dialogues. To tackle this challenge, the work proposes the first explainable agent architecture tailored for multi-category conversational fraud detection, integrating a summary-based memory mechanism with a dual-level (message- and conversation-level) detection framework that combines conversational semantic modeling and explainable AI techniques. The contributions include the release of ConScamBench-278, the first public multi-category benchmark dataset for conversational scams, achieving a 100% scam conversation detection rate on LoveFraud02 and 97.8% accuracy on ConScamBench-278. User studies further demonstrate that the system significantly enhances user trust and usability, attaining a usability score of 74.7.

0 citationsRead paper

DS-SAC: Density Search for Sample Consensus

Jul 04, 2026

This work addresses the overreliance on random sampling in robust geometric model estimation for computer vision by proposing a deterministic robust estimation framework. The method introduces, for the first time, deterministic dense-region search into the sample consensus paradigm, efficiently exploring high-consensus models through an initial-model-guided local forward-backward search combined with recursive partitioning based on signed residuals. This approach eliminates randomness while guaranteeing polynomial time complexity. Evaluated on homography, fundamental matrix, and essential matrix estimation tasks, the proposed method consistently outperforms mainstream approaches such as RANSAC and MAGSAC in terms of area under the cumulative error curve (AUC), median pose error, and computational speed.

0 citationsRead paper

Noise-Induced Landscape Distortion in QAOA for Constrained Binary Optimization: Empirical Characterization on IBM Quantum Hardware

Apr 21, 2026

Quantum hardware noise significantly flattens the variational energy landscape of the Quantum Approximate Optimization Algorithm (QAOA) for constrained binary optimization problems, thereby degrading algorithmic performance. This work proposes a device-agnostic metric—Landscape Span Compression (LSC)—to quantify noise-induced landscape distortion. Experiments on IBM’s ibm_fez processor demonstrate that LSC robustly characterizes noise severity compared to four existing metrics and effectively informs parameter transfer and error mitigation strategies. The study reveals that noise compresses the landscape span by 24–30% without shifting the optimal solution; feasible solutions at the optimal parameters remain 1.5–1.7 times more probable than random sampling; calibrated noise models account for only ~42% of the observed performance degradation; and zero-noise extrapolation yields limited gains while substantially increasing uncertainty.

0 citationsRead paper