Selected work

Representative Papers

ADABORD: a novel AdaBoost approach for ordinal classification

Jul 23, 2026

This work addresses the performance degradation in ordinal classification caused by existing methods' neglect of the natural order among classes. To this end, we propose ADABORD, a novel framework that, for the first time, integrates both an ordinal splitting criterion and an error function accounting for inter-class distances within AdaBoost. Specifically, ADABORD employs decision stumps based on an ordinal Gini impurity measure as base learners and introduces an absolute ranking probability score to more appropriately update sample and model weights. Experimental results on the TOC-UCO benchmark—the largest evaluation suite for ordinal classification—demonstrate that ADABORD significantly outperforms seven state-of-the-art methods, with particularly pronounced gains on datasets containing five or more ordinal classes.

0 citationsRead paper

Recursive ArUco Markers: A Scalable Fiducial Marker Design for Unmanned Aerial Vehicle Landing Pads

Jul 15, 2026

This work addresses the limitations of conventional ArUco markers, which suffer from poor detectability at extreme distances and high susceptibility to occlusion due to their reliance on a visible central region in recursive designs. To overcome these challenges, the authors propose a novel recursive ArUco marker that embeds complete child markers within the black-and-white bits of parent markers and employs an enhanced bit-sampling strategy. This design enables recursive nesting at arbitrary depths while ensuring consistent multi-scale detection, all without requiring visibility of the central region. The resulting marker maintains a single unique identifier and exhibits strong robustness against occlusion. Experimental results demonstrate that the proposed approach significantly extends the effective operational range and enhances occlusion resilience, offering a scalable, high-precision landing guidance solution for large-scale drone swarms.

0 citationsRead paper

MIHT: A Hoeffding Tree for Time Series Classification using Multiple Instance Learning

Mar 23, 2026

This work addresses the challenges of modeling and interpretation in multivariate, variable-length time series classification, where high dimensionality and heterogeneous sequence lengths complicate learning. The authors propose a multi-instance learning approach based on a “bag-of-subsequences” representation, which—by integrating Hoeffding trees with multi-instance learning for the first time—yields an incremental, white-box decision model. This method effectively discriminates between informative temporal segments and noise, automatically identifying salient variables and critical time intervals while maintaining model compactness and enhancing interpretability. Experimental evaluation on 28 public datasets demonstrates that the proposed approach outperforms 11 state-of-the-art algorithms in overall accuracy, with particularly strong performance in high-dimensional settings.

0 citationsRead paper

Behavioral Engagement in VR-Based Sign Language Learning: Visual Attention as a Predictor of Performance and Temporal Dynamics

Mar 19, 2026

This study addresses the effective measurement and prediction of learner engagement in virtual reality (VR)-based sign language learning and its impact on learning outcomes. Leveraging the VR sign language learning system SONAR, it introduces fine-grained visual attention trajectories as a core predictive variable for the first time. By integrating temporal dynamics and employing Pearson correlation, binomial generalized linear model (GLM) regression, and cross-user temporal aggregation of attention trajectories, the research investigates the relationship between engagement metrics and learning performance. Findings reveal that visual attention distribution and post-replay viewing duration significantly predict learning success, jointly accounting for a substantial proportion of variance in assessment scores. Moreover, the study uncovers attention peaks aligned with information density and phase-specific engagement dynamics, offering novel mechanisms and empirical support for modeling engagement in VR-based education.

0 citationsRead paper

Application of Time-Controlled Critical Point in Pressure Reducing Valves. A Case Study in North Spain

Dec 07, 2025

To address the challenge of imprecise pressure regulation at critical nodes in water distribution networks—particularly under scenarios with no flow signals or limited communication between critical nodes and pressure-reducing valves (PRVs), leading to persistently high leakage rates—this paper proposes a time-series-based collaborative pressure control method. Innovatively, it introduces the SARIMA model for the first time to dynamically model the nonlinear relationship between PRV outlet pressure and critical-node pressure, enabling feedforward pressure setpoint determination using only historical pressure data. Field validation in the Nohal region of Spain demonstrates statistical equivalence between predicted and measured pressures (p > 0.05), a 32% reduction in critical-node pressure fluctuations, and an 18.7% decrease in system leakage rate. The method establishes a low-cost, highly robust pressure optimization paradigm for aging infrastructure with constrained telemetry capabilities.

0 citationsRead paper

Recent publications

Latest Papers

ADABORD: a novel AdaBoost approach for ordinal classification

Jul 23, 2026

This work addresses the performance degradation in ordinal classification caused by existing methods' neglect of the natural order among classes. To this end, we propose ADABORD, a novel framework that, for the first time, integrates both an ordinal splitting criterion and an error function accounting for inter-class distances within AdaBoost. Specifically, ADABORD employs decision stumps based on an ordinal Gini impurity measure as base learners and introduces an absolute ranking probability score to more appropriately update sample and model weights. Experimental results on the TOC-UCO benchmark—the largest evaluation suite for ordinal classification—demonstrate that ADABORD significantly outperforms seven state-of-the-art methods, with particularly pronounced gains on datasets containing five or more ordinal classes.

0 citationsRead paper

Recursive ArUco Markers: A Scalable Fiducial Marker Design for Unmanned Aerial Vehicle Landing Pads

Jul 15, 2026

This work addresses the limitations of conventional ArUco markers, which suffer from poor detectability at extreme distances and high susceptibility to occlusion due to their reliance on a visible central region in recursive designs. To overcome these challenges, the authors propose a novel recursive ArUco marker that embeds complete child markers within the black-and-white bits of parent markers and employs an enhanced bit-sampling strategy. This design enables recursive nesting at arbitrary depths while ensuring consistent multi-scale detection, all without requiring visibility of the central region. The resulting marker maintains a single unique identifier and exhibits strong robustness against occlusion. Experimental results demonstrate that the proposed approach significantly extends the effective operational range and enhances occlusion resilience, offering a scalable, high-precision landing guidance solution for large-scale drone swarms.

0 citationsRead paper

MIHT: A Hoeffding Tree for Time Series Classification using Multiple Instance Learning

Mar 23, 2026

This work addresses the challenges of modeling and interpretation in multivariate, variable-length time series classification, where high dimensionality and heterogeneous sequence lengths complicate learning. The authors propose a multi-instance learning approach based on a “bag-of-subsequences” representation, which—by integrating Hoeffding trees with multi-instance learning for the first time—yields an incremental, white-box decision model. This method effectively discriminates between informative temporal segments and noise, automatically identifying salient variables and critical time intervals while maintaining model compactness and enhancing interpretability. Experimental evaluation on 28 public datasets demonstrates that the proposed approach outperforms 11 state-of-the-art algorithms in overall accuracy, with particularly strong performance in high-dimensional settings.

0 citationsRead paper

Behavioral Engagement in VR-Based Sign Language Learning: Visual Attention as a Predictor of Performance and Temporal Dynamics

Mar 19, 2026

This study addresses the effective measurement and prediction of learner engagement in virtual reality (VR)-based sign language learning and its impact on learning outcomes. Leveraging the VR sign language learning system SONAR, it introduces fine-grained visual attention trajectories as a core predictive variable for the first time. By integrating temporal dynamics and employing Pearson correlation, binomial generalized linear model (GLM) regression, and cross-user temporal aggregation of attention trajectories, the research investigates the relationship between engagement metrics and learning performance. Findings reveal that visual attention distribution and post-replay viewing duration significantly predict learning success, jointly accounting for a substantial proportion of variance in assessment scores. Moreover, the study uncovers attention peaks aligned with information density and phase-specific engagement dynamics, offering novel mechanisms and empirical support for modeling engagement in VR-based education.

0 citationsRead paper

Application of Time-Controlled Critical Point in Pressure Reducing Valves. A Case Study in North Spain

Dec 07, 2025

To address the challenge of imprecise pressure regulation at critical nodes in water distribution networks—particularly under scenarios with no flow signals or limited communication between critical nodes and pressure-reducing valves (PRVs), leading to persistently high leakage rates—this paper proposes a time-series-based collaborative pressure control method. Innovatively, it introduces the SARIMA model for the first time to dynamically model the nonlinear relationship between PRV outlet pressure and critical-node pressure, enabling feedforward pressure setpoint determination using only historical pressure data. Field validation in the Nohal region of Spain demonstrates statistical equivalence between predicted and measured pressures (p > 0.05), a 32% reduction in critical-node pressure fluctuations, and an 18.7% decrease in system leakage rate. The method establishes a low-cost, highly robust pressure optimization paradigm for aging infrastructure with constrained telemetry capabilities.

0 citationsRead paper