Selected work

Representative Papers

Flexible Coupler Antenna Enhanced Wireless Communication: Modeling and Coupler Position Optimization

Jun 07, 2026

This study addresses the high complexity and cost of conventional active array beamforming by proposing a low-power alternative based on a flexible coupler-based antenna architecture. The design enables mechanical beam steering through physical displacement of passive coupling elements alone, modulating induced currents without requiring any adjustment to active antennas. This approach pioneers purely passive element repositioning for beam control, substantially reducing both hardware cost and power consumption. Leveraging multi-port circuit theory, the authors develop line-of-sight and multipath channel models and employ a block coordinate conditional gradient algorithm to optimize coupler placement. Experimental results demonstrate that, despite significantly fewer active elements and RF chains, the proposed system achieves notably higher spectral efficiency compared to existing benchmark schemes.

0 citationsRead paper

Outlier detection in mixed-attribute data: a semi-supervised approach with fuzzy approximations and relative entropy

Dec 21, 2025

To address the challenges of uncertainty and heterogeneity in anomaly detection for mixed-attribute data, this paper proposes FROD, a semi-supervised framework. FROD is the first method to jointly leverage fuzzy rough set modeling and fuzzy relative entropy quantification: it employs a small set of labeled instances to assess attribute discriminability, and computes anomaly scores via a synergistic measure—fuzzy approximation accuracy and relative entropy—derived from unlabeled data. By explicitly modeling both uncertainty and structural heterogeneity inherent in mixed-attribute spaces, FROD relaxes the strong homogeneity and determinism assumptions underlying conventional approaches. Extensive experiments across 16 public benchmark datasets demonstrate that FROD matches or surpasses state-of-the-art methods in detection performance. All code and datasets are publicly released, confirming its robustness and effectiveness in real-world scenarios.

0 citationsRead paper

Label-Informed Outlier Detection Based on Granule Density

Dec 21, 2025

Existing semi-supervised anomaly detection methods for heterogeneous data overlook data heterogeneity and uncertainty. To address this, we propose Label-Guided Granular Density Outlier Factor (GDOF), the first approach to embed sparse anomaly labels into a fuzzy granulation process. GDOF constructs an attribute-adaptive granular density ensemble: it models multi-granularity uncertainty via fuzzy sets, captures heterogeneous attribute structures using granular computing principles, and enhances discriminability through label-guided density estimation and attribute-correlation-weighted fusion. Extensive experiments on multiple real-world heterogeneous datasets demonstrate that GDOF achieves state-of-the-art performance with only a minimal number of labeled anomalies (e.g., 5–10 samples), significantly outperforming existing semi-supervised methods.

0 citationsRead paper

Robust Duality Learning for Unsupervised Visible-Infrared Person Re-Identification

May 05, 2025IEEE Transactions on Information Forensics and Security

Unsupervised visible-infrared person re-identification (UVI-ReID) confronts dual challenges: cross-modal heterogeneity and pseudo-label noise—including noise overfitting, error accumulation, and inter-cluster mismatch. To address these, we propose RoDE, a robust dual-learning framework that innovatively integrates adaptive robust anti-noise learning (RAL), alternating dual training, and cluster-consistent matching (CCM). RoDE explicitly models and suppresses pseudo-label noise via dynamic sample reweighting, alternating self-training between two complementary models, and similarity-driven cross-modal cluster alignment. Extensive experiments on SYSU-MM01, RegDB, and LLVIP benchmarks demonstrate state-of-the-art performance, with mAP improvements of up to 6.2% over prior methods. Ablation studies confirm the effectiveness of each component in enhancing noise robustness and cross-modal generalization.

0 citationsRead paper

Deep Reversible Consistency Learning for Cross-modal Retrieval

Jan 10, 2025

To address key bottlenecks in cross-modal retrieval—including strong inter-modal coupling, weak semantic alignment, and blind prior selection—this paper proposes the Deep Reversible Consistency Learning (DRCL) framework. DRCL introduces two novel components: Selective Prior Learning (SPL), which adaptively identifies modality-agnostic priors, and Reversible Semantic Consistency (RSC) learning, which employs generalized matrix inverses to enable invertible mapping from labels to disentangled representations. Furthermore, DRCL integrates modality-invariance guidance and feature enhancement to improve distributional robustness. Critically, the method supports training without paired samples, thereby mitigating spurious inter-modal correlation assumptions. Extensive experiments on five benchmark datasets demonstrate that DRCL consistently outperforms 15 state-of-the-art methods, achieving significant gains in both retrieval accuracy and generalization across diverse cross-modal settings.

0 citationsRead paper

Recent publications

Latest Papers

Flexible Coupler Antenna Enhanced Wireless Communication: Modeling and Coupler Position Optimization

Jun 07, 2026

This study addresses the high complexity and cost of conventional active array beamforming by proposing a low-power alternative based on a flexible coupler-based antenna architecture. The design enables mechanical beam steering through physical displacement of passive coupling elements alone, modulating induced currents without requiring any adjustment to active antennas. This approach pioneers purely passive element repositioning for beam control, substantially reducing both hardware cost and power consumption. Leveraging multi-port circuit theory, the authors develop line-of-sight and multipath channel models and employ a block coordinate conditional gradient algorithm to optimize coupler placement. Experimental results demonstrate that, despite significantly fewer active elements and RF chains, the proposed system achieves notably higher spectral efficiency compared to existing benchmark schemes.

0 citationsRead paper

Outlier detection in mixed-attribute data: a semi-supervised approach with fuzzy approximations and relative entropy

Dec 21, 2025

To address the challenges of uncertainty and heterogeneity in anomaly detection for mixed-attribute data, this paper proposes FROD, a semi-supervised framework. FROD is the first method to jointly leverage fuzzy rough set modeling and fuzzy relative entropy quantification: it employs a small set of labeled instances to assess attribute discriminability, and computes anomaly scores via a synergistic measure—fuzzy approximation accuracy and relative entropy—derived from unlabeled data. By explicitly modeling both uncertainty and structural heterogeneity inherent in mixed-attribute spaces, FROD relaxes the strong homogeneity and determinism assumptions underlying conventional approaches. Extensive experiments across 16 public benchmark datasets demonstrate that FROD matches or surpasses state-of-the-art methods in detection performance. All code and datasets are publicly released, confirming its robustness and effectiveness in real-world scenarios.

0 citationsRead paper

Label-Informed Outlier Detection Based on Granule Density

Dec 21, 2025

Existing semi-supervised anomaly detection methods for heterogeneous data overlook data heterogeneity and uncertainty. To address this, we propose Label-Guided Granular Density Outlier Factor (GDOF), the first approach to embed sparse anomaly labels into a fuzzy granulation process. GDOF constructs an attribute-adaptive granular density ensemble: it models multi-granularity uncertainty via fuzzy sets, captures heterogeneous attribute structures using granular computing principles, and enhances discriminability through label-guided density estimation and attribute-correlation-weighted fusion. Extensive experiments on multiple real-world heterogeneous datasets demonstrate that GDOF achieves state-of-the-art performance with only a minimal number of labeled anomalies (e.g., 5–10 samples), significantly outperforming existing semi-supervised methods.

0 citationsRead paper

Robust Duality Learning for Unsupervised Visible-Infrared Person Re-Identification

May 05, 2025IEEE Transactions on Information Forensics and Security

Unsupervised visible-infrared person re-identification (UVI-ReID) confronts dual challenges: cross-modal heterogeneity and pseudo-label noise—including noise overfitting, error accumulation, and inter-cluster mismatch. To address these, we propose RoDE, a robust dual-learning framework that innovatively integrates adaptive robust anti-noise learning (RAL), alternating dual training, and cluster-consistent matching (CCM). RoDE explicitly models and suppresses pseudo-label noise via dynamic sample reweighting, alternating self-training between two complementary models, and similarity-driven cross-modal cluster alignment. Extensive experiments on SYSU-MM01, RegDB, and LLVIP benchmarks demonstrate state-of-the-art performance, with mAP improvements of up to 6.2% over prior methods. Ablation studies confirm the effectiveness of each component in enhancing noise robustness and cross-modal generalization.

0 citationsRead paper

Deep Reversible Consistency Learning for Cross-modal Retrieval

Jan 10, 2025

To address key bottlenecks in cross-modal retrieval—including strong inter-modal coupling, weak semantic alignment, and blind prior selection—this paper proposes the Deep Reversible Consistency Learning (DRCL) framework. DRCL introduces two novel components: Selective Prior Learning (SPL), which adaptively identifies modality-agnostic priors, and Reversible Semantic Consistency (RSC) learning, which employs generalized matrix inverses to enable invertible mapping from labels to disentangled representations. Furthermore, DRCL integrates modality-invariance guidance and feature enhancement to improve distributional robustness. Critically, the method supports training without paired samples, thereby mitigating spurious inter-modal correlation assumptions. Extensive experiments on five benchmark datasets demonstrate that DRCL consistently outperforms 15 state-of-the-art methods, achieving significant gains in both retrieval accuracy and generalization across diverse cross-modal settings.

0 citationsRead paper