Selected work

Representative Papers

ProtoHGF-Net: Prototype HyperGraph Fusion with Intra-modal Calibration for RGBT Object Detection

Aug 11, 2026

This work addresses the challenge in existing RGBT object detection methods, where dense cross-modal interactions are prone to background interference and struggle to focus on target semantics. To overcome this limitation, the authors propose a Prototype Hypergraph Fusion Network that reformulates cross-modal fusion as prototype-level semantic interaction. Additionally, they introduce a teacher-mask calibration distillation strategy to perform target-aware calibration of modality-specific features prior to fusion, effectively suppressing background noise and enhancing target representation. The proposed method achieves state-of-the-art performance with mAP50 scores of 85.9%, 88.2%, and 79.1% on the DroneVehicle, DVTOD, and FLIR datasets, respectively.

0 citationsRead paper

Privileged Lesion-Context Relational Distillation for Mask-Free Skin Lesion Classification

Jul 21, 2026

This work addresses the challenge of skin lesion classification, which often relies on segmentation masks or auxiliary models that hinder clinical deployment and increase computational overhead. The authors propose the PLCRD framework, which leverages lesion masks during training to construct a teacher model and transfers structured knowledge about lesion–context relationships to a student model that requires only raw images for inference, thereby enabling mask-free prediction. Innovatively, privileged mask information is transformed into transferable relational knowledge, circumventing direct feature alignment between heterogeneous architectures. The approach integrates multiple mechanisms—including diagnostic distribution transfer, attention propagation, lesion similarity alignment, and lesion–context affinity matching. Evaluated on HAM10000 and ISIC 2018, the method achieves macro F1 scores of 0.773 ± 0.018 and 0.732 ± 0.008, respectively, significantly advancing classification performance under mask-free conditions.

0 citationsRead paper

Beyond Weather Correlation: A Comparative Study of Static and Temporal Neural Architectures for Fine-Grained Residential Energy Consumption Forecasting in Melbourne, Australia

Apr 14, 2026

This study addresses the challenge of 5-minute resolution electricity consumption forecasting for Australian households by integrating temporal autocorrelation and static meteorological features, with a comparative evaluation of MLP and LSTM models. Leveraging 14 months of smart meter data from two Melbourne households alongside weather observations, the analysis employs a 24-step sliding window, fuses meteorological variables, and incorporates seasonal stratification. The work is the first to quantify the dominant role of temporal autocorrelation in high-granularity residential load forecasting and uncovers the implicit influence of weather on photovoltaic generation in solar-equipped homes. Experimental results demonstrate that LSTM substantially outperforms MLP, achieving R² scores of 0.883 and 0.865 compared to MLP’s −0.055 and 0.410, thereby confirming the critical advantage of explicit temporal modeling in fine-grained electricity demand prediction.

0 citationsRead paper

A Monad-Based Clause Architecture for Artificial Age Score (AAS) in Large Language Models

Dec 03, 2025

Large language models (LLMs) lack auditable, enforceable governance mechanisms for the evolution of internal memory. Method: We propose a novel, Leibnizian monad-inspired architecture comprising six executable clauses—formalizing monadology as a code-level regulatory system—and introduce kernelized Artificial Memory Aging Scoring (AAS) to drive memory updates, underpinned by three core mechanisms: continuity assurance, contradiction penalization, and hierarchical interpretability. Contribution/Results: Evaluated via Monad-semantic modeling, Python-based numerical experiments, and channel-level metrics (recall, redundancy, weight evolution) alongside windowed Perfectness Drift Detection, our framework ensures controllable, continuous memory trajectories; explicitly suppresses contradictory or unsupported assertions; enables organic, traceable hierarchical structure; achieves dual-perspective (model/user) alignment; and reliably distinguishes sustained model improvement from performance degradation.

0 citationsRead paper

Enhancing Bankruptcy Prediction of Banks through Advanced Machine Learning Techniques: An Innovative Approach and Analysis

Oct 08, 2025

This study addresses the limitations of conventional statistical models—such as restrictive linearity and distributional assumptions—and insufficient predictive accuracy in bank failure forecasting. To overcome these challenges, we propose a machine learning–based analytical framework leveraging multi-source financial data, incorporating logistic regression, random forest, and support vector machine algorithms. The models are empirically validated on commercial and rural bank datasets from Turkey and Indonesia. Random forest achieves 90% classification accuracy on commercial banks, while all three models consistently identify failure trends among rural banks. Our key contribution lies in relaxing strong parametric assumptions, thereby enhancing model generalizability across heterogeneous bank types and geographically distinct regions. The framework delivers a robust, empirically validated, and scalable technical approach for early systemic risk detection and evidence-based regulatory decision-making.

0 citationsRead paper

Recent publications

Latest Papers

ProtoHGF-Net: Prototype HyperGraph Fusion with Intra-modal Calibration for RGBT Object Detection

Aug 11, 2026

This work addresses the challenge in existing RGBT object detection methods, where dense cross-modal interactions are prone to background interference and struggle to focus on target semantics. To overcome this limitation, the authors propose a Prototype Hypergraph Fusion Network that reformulates cross-modal fusion as prototype-level semantic interaction. Additionally, they introduce a teacher-mask calibration distillation strategy to perform target-aware calibration of modality-specific features prior to fusion, effectively suppressing background noise and enhancing target representation. The proposed method achieves state-of-the-art performance with mAP50 scores of 85.9%, 88.2%, and 79.1% on the DroneVehicle, DVTOD, and FLIR datasets, respectively.

0 citationsRead paper

Privileged Lesion-Context Relational Distillation for Mask-Free Skin Lesion Classification

Jul 21, 2026

This work addresses the challenge of skin lesion classification, which often relies on segmentation masks or auxiliary models that hinder clinical deployment and increase computational overhead. The authors propose the PLCRD framework, which leverages lesion masks during training to construct a teacher model and transfers structured knowledge about lesion–context relationships to a student model that requires only raw images for inference, thereby enabling mask-free prediction. Innovatively, privileged mask information is transformed into transferable relational knowledge, circumventing direct feature alignment between heterogeneous architectures. The approach integrates multiple mechanisms—including diagnostic distribution transfer, attention propagation, lesion similarity alignment, and lesion–context affinity matching. Evaluated on HAM10000 and ISIC 2018, the method achieves macro F1 scores of 0.773 ± 0.018 and 0.732 ± 0.008, respectively, significantly advancing classification performance under mask-free conditions.

0 citationsRead paper

Beyond Weather Correlation: A Comparative Study of Static and Temporal Neural Architectures for Fine-Grained Residential Energy Consumption Forecasting in Melbourne, Australia

Apr 14, 2026

This study addresses the challenge of 5-minute resolution electricity consumption forecasting for Australian households by integrating temporal autocorrelation and static meteorological features, with a comparative evaluation of MLP and LSTM models. Leveraging 14 months of smart meter data from two Melbourne households alongside weather observations, the analysis employs a 24-step sliding window, fuses meteorological variables, and incorporates seasonal stratification. The work is the first to quantify the dominant role of temporal autocorrelation in high-granularity residential load forecasting and uncovers the implicit influence of weather on photovoltaic generation in solar-equipped homes. Experimental results demonstrate that LSTM substantially outperforms MLP, achieving R² scores of 0.883 and 0.865 compared to MLP’s −0.055 and 0.410, thereby confirming the critical advantage of explicit temporal modeling in fine-grained electricity demand prediction.

0 citationsRead paper

A Monad-Based Clause Architecture for Artificial Age Score (AAS) in Large Language Models

Dec 03, 2025

Large language models (LLMs) lack auditable, enforceable governance mechanisms for the evolution of internal memory. Method: We propose a novel, Leibnizian monad-inspired architecture comprising six executable clauses—formalizing monadology as a code-level regulatory system—and introduce kernelized Artificial Memory Aging Scoring (AAS) to drive memory updates, underpinned by three core mechanisms: continuity assurance, contradiction penalization, and hierarchical interpretability. Contribution/Results: Evaluated via Monad-semantic modeling, Python-based numerical experiments, and channel-level metrics (recall, redundancy, weight evolution) alongside windowed Perfectness Drift Detection, our framework ensures controllable, continuous memory trajectories; explicitly suppresses contradictory or unsupported assertions; enables organic, traceable hierarchical structure; achieves dual-perspective (model/user) alignment; and reliably distinguishes sustained model improvement from performance degradation.

0 citationsRead paper

Enhancing Bankruptcy Prediction of Banks through Advanced Machine Learning Techniques: An Innovative Approach and Analysis

Oct 08, 2025

This study addresses the limitations of conventional statistical models—such as restrictive linearity and distributional assumptions—and insufficient predictive accuracy in bank failure forecasting. To overcome these challenges, we propose a machine learning–based analytical framework leveraging multi-source financial data, incorporating logistic regression, random forest, and support vector machine algorithms. The models are empirically validated on commercial and rural bank datasets from Turkey and Indonesia. Random forest achieves 90% classification accuracy on commercial banks, while all three models consistently identify failure trends among rural banks. Our key contribution lies in relaxing strong parametric assumptions, thereby enhancing model generalizability across heterogeneous bank types and geographically distinct regions. The framework delivers a robust, empirically validated, and scalable technical approach for early systemic risk detection and evidence-based regulatory decision-making.

0 citationsRead paper