Xiaona Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Xiaona Liu has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Difan Zou
The University of Hong Kong
2 shared publications - Yang Wang
The Institute of IoT, Shenzhen Polytechnic University, Shenzhen 518055, China
2 shared publications - Difei Wu
Key Laboratory of Road and Traffic Engineering of the Ministry of Education, Tongji University, Shanghai 201804, China
1 shared publication - Le Zheng
School of Information and Electronics, Beijing Institute of Technology, Beijing, China
1 shared publication Shilin Tong
1 shared publication- Yang Wang
Shenzhen Polytech, IoT Res Inst, Shenzhen 518055, Peoples R China
1 shared publication
Latest Publications
- Open Access
- 12 September 2026
- Published by MDPI AG in Electronics
- Vol. 15 (18) , 4140
Bayesian Receiver Orientation Steering for Indoor Visible Light Positioning
Received-signal-strength (RSS) visible light positioning (VLP) with a single photodiode (PD) usually assumes an upward-facing receiver. This convention simplifies the channel model, but it leaves receiver orientation unused even though orientation changes RSS sensitivity and LED visibility. We study a two-measurement localization cycle: an initial upward RSS vector gives a coarse position estimate, and we then orient the PD before acquiring a second vector. We use expected posterior localization mean-square error to motivate the sensing decision and develop two efficient local-information policies. Point scores the coarse estimate, whereas Stencil examines a nine-point neighborhood while enforcing LED visibility. After steering, we jointly fit the upward and steered RSS vectors for every policy. Using common Monte Carlo realizations and paired bootstrap intervals, we compare the policies with fixed upward reception, random steering, center pointing, and maximum-predicted-RSS steering. In 4-LED and 6-LED rooms, both local-information policies reduce RMSE by 18–20% relative to fixed upward reception; in an 8 m room, the reductions are about 40%. We further compare the selected actions with a finite grid-posterior reference and examine sensitivity to stencil parameters and actuator rotation error. The reference confirms that Point and Stencil are Bayes-risk-motivated surrogates, not exact Bayes optimizers.
- 12 September 2026
- Published by MDPI AG in Electronics
- Vol. 15 (18) , 4140
Open AccessBayesian Receiver Orientation Steering for Indoor Visible Light Positioning
Received-signal-strength (RSS) visible light positioning (VLP) with a single photodiode (PD) usually assumes an upward-facing receiver. This convention simplifies the channel model, but it leaves receiver orientation unused even though orientation changes RSS sensitivity and LED visibility. We study a two-measurement localization cycle: an initial upward RSS vector gives a coarse position estimate, and we then orient the PD before acquiring a second vector. We use expected posterior localization mean-square error to motivate the sensing decision and develop two efficient local-information policies. Point scores the coarse estimate, whereas Stencil examines a nine-point neighborhood while enforcing LED visibility. After steering, we jointly fit the upward and steered RSS vectors for every policy. Using common Monte Carlo realizations and paired bootstrap intervals, we compare the policies with fixed upward reception, random steering, center pointing, and maximum-predicted-RSS steering. In 4-LED and 6-LED rooms, both local-information policies reduce RMSE by 18–20% relative to fixed upward reception; in an 8 m room, the reductions are about 40%. We further compare the selected actions with a finite grid-posterior reference and examine sensitivity to stencil parameters and actuator rotation error. The reference confirms that Point and Stencil are Bayes-risk-motivated surrogates, not exact Bayes optimizers.
- Open Access
- 19 July 2026
- Published by MDPI AG in Electronics
- Vol. 15 (14) , 3172
Efficient 3D Indoor Visible Light Positioning via Stabilized Regularized Least Squares
We study indoor visible light positioning (VLP) in a fully three-dimensional setting under thermal noise and signal-dependent shot noise. Existing shot-noise-aware least-squares (LS) VLP methods largely rely on a fixed receiver height, so their key simplifications break down when the vertical coordinate is unknown. We derive a 3D likelihood-inspired surrogate, lift the nonlinear geometry into a weighted LS form, and add a geometry-aware regularizer that restores approximate consistency among redundant lifted variables. To make the iterative solver reliable in sparse layouts, we introduce room-feasible projection, consistent pilot reconstruction, clipped frozen weights, and damped objective-accepted updates. Each inner step remains a closed-form 4 × 4 linear solve. In deterministic synthetic simulations of an 8 × 8 × 3 m room with layouts containing 4, 9, and 16 light-emitting diodes (LEDs), high- and low-power regimes, and 400 test samples per scenario, the stabilized solver removes the catastrophic sparse-layout failures observed for the tested LS-family baselines and retains a 14×–23× runtime advantage over the nonlinear baseline. The nonlinear baseline often attains lower mean error in sparse and moderate layouts; the contribution here is the resulting speed–tail-control trade-off within the fixed-orientation line-of-sight model. The resulting estimator is intended for synthetic 3D VLP simulations; hardware validation is outside the present study.
- 19 July 2026
- Published by MDPI AG in Electronics
- Vol. 15 (14) , 3172
Open AccessEfficient 3D Indoor Visible Light Positioning via Stabilized Regularized Least Squares
We study indoor visible light positioning (VLP) in a fully three-dimensional setting under thermal noise and signal-dependent shot noise. Existing shot-noise-aware least-squares (LS) VLP methods largely rely on a fixed receiver height, so their key simplifications break down when the vertical coordinate is unknown. We derive a 3D likelihood-inspired surrogate, lift the nonlinear geometry into a weighted LS form, and add a geometry-aware regularizer that restores approximate consistency among redundant lifted variables. To make the iterative solver reliable in sparse layouts, we introduce room-feasible projection, consistent pilot reconstruction, clipped frozen weights, and damped objective-accepted updates. Each inner step remains a closed-form 4 × 4 linear solve. In deterministic synthetic simulations of an 8 × 8 × 3 m room with layouts containing 4, 9, and 16 light-emitting diodes (LEDs), high- and low-power regimes, and 400 test samples per scenario, the stabilized solver removes the catastrophic sparse-layout failures observed for the tested LS-family baselines and retains a 14×–23× runtime advantage over the nonlinear baseline. The nonlinear baseline often attains lower mean error in sparse and moderate layouts; the contribution here is the resulting speed–tail-control trade-off within the fixed-orientation line-of-sight model. The resulting estimator is intended for synthetic 3D VLP simulations; hardware validation is outside the present study.
- Open Access
- 27 May 2026
- Published by MDPI AG in Electronics
- Vol. 15 (11) , 2316
Wasserstein-Robust Training for One-Hidden-Layer ReLU Networks with Distributional Guarantees
Neural networks are vulnerable to adversarial perturbations, which motivates training procedures with formal robustness guarantees. In this paper, we study one-hidden-layer ReLU networks from the perspective of Wasserstein distributional robustness. Leveraging the network structure, we derive an upper bound for the intractable robust surrogate in the form of a tractable regularized empirical risk objective whose regularizer is computed through a low-rank optimization problem based on Burer–Monteiro factorization. This reformulation yields a distributional robustness certificate on the worst-case expected loss over a Wasserstein ball. The upper bound construction and distributional certificate are developed for the shallow fixed-output multiclass formulation, while the optimization analysis focuses on a binary specialization with margin loss and exact linear separability. We also analyze a modified stochastic gradient descent scheme for the resulting regularized problem in this binary linearly separable setting, and we establish a corresponding generalization bound. The experiments validate the proposed surrogate and training procedure on binary MNIST and CIFAR-10 tasks, and we added a 10-class MNIST experiment to further check the multiclass trainability of the surrogate.
- 27 May 2026
- Published by MDPI AG in Electronics
- Vol. 15 (11) , 2316
Open AccessWasserstein-Robust Training for One-Hidden-Layer ReLU Networks with Distributional Guarantees
Neural networks are vulnerable to adversarial perturbations, which motivates training procedures with formal robustness guarantees. In this paper, we study one-hidden-layer ReLU networks from the perspective of Wasserstein distributional robustness. Leveraging the network structure, we derive an upper bound for the intractable robust surrogate in the form of a tractable regularized empirical risk objective whose regularizer is computed through a low-rank optimization problem based on Burer–Monteiro factorization. This reformulation yields a distributional robustness certificate on the worst-case expected loss over a Wasserstein ball. The upper bound construction and distributional certificate are developed for the shallow fixed-output multiclass formulation, while the optimization analysis focuses on a binary specialization with margin loss and exact linear separability. We also analyze a modified stochastic gradient descent scheme for the resulting regularized problem in this binary linearly separable setting, and we establish a corresponding generalization bound. The experiments validate the proposed surrogate and training procedure on binary MNIST and CIFAR-10 tasks, and we added a 10-class MNIST experiment to further check the multiclass trainability of the surrogate.
- Open Access
STGAN: Spatial-temporal Graph Autoregression Network for Pavement Distress Deterioration Prediction
Pavement distress significantly compromises road integrity and poses risks to drivers. Accurate prediction of pavement distress deterioration is essential for effective road management, cost reduction in maintenance, and improvement of traffic safety. However, real-world data on pavement distress is usually collected irregularly, resulting in uneven, asynchronous, and sparse spatial-temporal datasets. This hinders the application of existing spatial-temporal models, such as DCRNN, since they are only applicable to regularly and synchronously collected data. To overcome these challenges, we propose the Spatial-Temporal Graph Autoregression Network (STGAN), a novel graph neural network model designed for accurately predicting irregular pavement distress deterioration using complex spatial-temporal data. Specifically, STGAN integrates the temporal domain into the spatial domain, creating a larger graph where nodes are represented by spatial-temporal tuples and edges are formed based on a similarity-based connection mechanism. Furthermore, based on the constructed spatiotemporal graph, we formulate pavement distress deterioration prediction as a graph autoregression task, i.e., the graph size increases incrementally and the prediction is performed sequentially. This is accomplished by a novel spatial-temporal attention mechanism deployed by STGAN. Utilizing the ConTrack dataset, which contains pavement distress records collected from different locations in Shanghai, we demonstrate the superior performance of STGAN in capturing spatial-temporal correlations and addressing the aforementioned challenges. Experimental results further show that STGAN outperforms baseline models, and ablation studies confirm the effectiveness of its novel modules. Our findings contribute to promoting proactive road maintenance decision-making and ultimately enhancing road safety and resilience.
Open AccessSTGAN: Spatial-temporal Graph Autoregression Network for Pavement Distress Deterioration Prediction
Pavement distress significantly compromises road integrity and poses risks to drivers. Accurate prediction of pavement distress deterioration is essential for effective road management, cost reduction in maintenance, and improvement of traffic safety. However, real-world data on pavement distress is usually collected irregularly, resulting in uneven, asynchronous, and sparse spatial-temporal datasets. This hinders the application of existing spatial-temporal models, such as DCRNN, since they are only applicable to regularly and synchronously collected data. To overcome these challenges, we propose the Spatial-Temporal Graph Autoregression Network (STGAN), a novel graph neural network model designed for accurately predicting irregular pavement distress deterioration using complex spatial-temporal data. Specifically, STGAN integrates the temporal domain into the spatial domain, creating a larger graph where nodes are represented by spatial-temporal tuples and edges are formed based on a similarity-based connection mechanism. Furthermore, based on the constructed spatiotemporal graph, we formulate pavement distress deterioration prediction as a graph autoregression task, i.e., the graph size increases incrementally and the prediction is performed sequentially. This is accomplished by a novel spatial-temporal attention mechanism deployed by STGAN. Utilizing the ConTrack dataset, which contains pavement distress records collected from different locations in Shanghai, we demonstrate the superior performance of STGAN in capturing spatial-temporal correlations and addressing the aforementioned challenges. Experimental results further show that STGAN outperforms baseline models, and ablation studies confirm the effectiveness of its novel modules. Our findings contribute to promoting proactive road maintenance decision-making and ultimately enhancing road safety and resilience.