Nguyen Nang Hung Van
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Nguyen Nang Hung Van has a total of 30 co-authors. Here are the top co-authors by number of shared publications:
- Phuc Hao Do
Danang Architecture University
9 shared publications - Van Nam Hoang
Dong A University
5 shared publications - Tran Duc Le
Department of Mathematics, Statistics & Computer Science, University of Wisconsin-Stout, Menomonie, WI 54751, USA
4 shared publications Truc Thi Kim Nguyen
3 shared publications- Hiep Xuan Huynh
Can Tho University
2 shared publications - Le Thi My Hanh
The University of Da Nang, University of Science and Technology, Danang, Vietnam
2 shared publications
Latest Publications
- 30 September 2026
- Published by Springer Nature
- p. 198-209
MC-JEPA: A Neural Network Framework with Multi-criteria Aggregation for Recommender Systems
- 30 September 2026
- Published by Springer Nature
- p. 198-209
MC-JEPA: A Neural Network Framework with Multi-criteria Aggregation for Recommender Systems
- 2 September 2026
- Published by Springer Nature
- p. 139-150
Polynomial-Augmented Instant Neural Graphics Primitives
- 2 September 2026
- Published by Springer Nature
- p. 139-150
Polynomial-Augmented Instant Neural Graphics Primitives
- 2 August 2026
- Published by Springer Nature
- p. 1-12
Enhancing Medical Image Classification with Noise-Injected Multi-Head Attention
- 2 August 2026
- Published by Springer Nature
- p. 1-12
Enhancing Medical Image Classification with Noise-Injected Multi-Head Attention
- Open Access
- 29 May 2026
- Published by Elsevier in SSRN Electronic Journal
Architectural ablation and statistical quality control for deep stellar spectral classification on SDSS DR19
The Sloan Digital Sky Survey (SDSS) Data Release~19 provides oneof the largest public collections of optical stellar spectra, yetpipeline quality flags alone do not guarantee clean inputs fordeep-learning classifiers, and the optimal architecture forone-dimensional spectral data remains open.We present an end-to-end pipeline for seven-class main-sequencespectral typing (O--M) that combines (i)~a strict upstreamquality cut ($\snr > 25$, \texttt{zWarning}=0, luminosity class V),(ii)~wavelength resampling and continuum normalisation tofixed-length $(1,4000)$ vectors, and (iii)~a \emph{statisticalpost-processing quality-control} layer that flags continuum-spikeartefacts and corrupted spectra via percentile thresholds on theper-spectrum mean and standard deviation, removing $1{,}814$spectra (1.05\,\%). A \emph{smoothed class-weighting} scheme$W_{\mathrm{class}} = W_{\mathrm{balanced}}^{0.7}$ mitigates theclass imbalance amplified by the $\snr$ cut, yielding a markedlybetter precision/recall trade-off than standard balanced weights.Under matched conditions we ablate four architectures: a Pure1D-CNN, a CNN + Bi-LSTM hybrid, a shallow Conv-Transformer, and aDeep CNN-Transformer with positional encoding.On a stratified test split the Pure 1D-CNN achieves 92.90\,\%accuracy, outperforming all three hybrids at the lowest trainingcost. The two Conv-Transformer variants isolate positionalencoding: removal costs over eight accuracy points. Remaining errorsconcentrate on the F--G boundary, a physically grounded limitdriven by gradual neutral-metal line variation; self-attentionsmooths the narrow peaks that disambiguate this regime.Careful data curation and local feature extraction thereforematter more than architectural complexity for stellar spectroscopy;we release the QC tooling and reference implementations to supportreproducible benchmarking.- 29 May 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessArchitectural ablation and statistical quality control for deep stellar spectral classification on SDSS DR19
The Sloan Digital Sky Survey (SDSS) Data Release~19 provides oneof the largest public collections of optical stellar spectra, yetpipeline quality flags alone do not guarantee clean inputs fordeep-learning classifiers, and the optimal architecture forone-dimensional spectral data remains open.We present an end-to-end pipeline for seven-class main-sequencespectral typing (O--M) that combines (i)~a strict upstreamquality cut ($\snr > 25$, \texttt{zWarning}=0, luminosity class V),(ii)~wavelength resampling and continuum normalisation tofixed-length $(1,4000)$ vectors, and (iii)~a \emph{statisticalpost-processing quality-control} layer that flags continuum-spikeartefacts and corrupted spectra via percentile thresholds on theper-spectrum mean and standard deviation, removing $1{,}814$spectra (1.05\,\%). A \emph{smoothed class-weighting} scheme$W_{\mathrm{class}} = W_{\mathrm{balanced}}^{0.7}$ mitigates theclass imbalance amplified by the $\snr$ cut, yielding a markedlybetter precision/recall trade-off than standard balanced weights.Under matched conditions we ablate four architectures: a Pure1D-CNN, a CNN + Bi-LSTM hybrid, a shallow Conv-Transformer, and aDeep CNN-Transformer with positional encoding.On a stratified test split the Pure 1D-CNN achieves 92.90\,\%accuracy, outperforming all three hybrids at the lowest trainingcost. The two Conv-Transformer variants isolate positionalencoding: removal costs over eight accuracy points. Remaining errorsconcentrate on the F--G boundary, a physically grounded limitdriven by gradual neutral-metal line variation; self-attentionsmooths the narrow peaks that disambiguate this regime.Careful data curation and local feature extraction thereforematter more than architectural complexity for stellar spectroscopy;we release the QC tooling and reference implementations to supportreproducible benchmarking. - 3 April 2026
- Published by Springer Nature
- p. 598-612
Enhancing Cross-Project Test Smell Detection via Test Code Feature Engineering and Ensemble Learning
- 3 April 2026
- Published by Springer Nature
- p. 598-612
Enhancing Cross-Project Test Smell Detection via Test Code Feature Engineering and Ensemble Learning
- 1 April 2026
- Published by Springer Nature
- p. 156-164
Robust Quishing Detection on Low-Resolution QR Codes via Feature Learning and Residual CNNs
- 1 April 2026
- Published by Springer Nature
- p. 156-164
Robust Quishing Detection on Low-Resolution QR Codes via Feature Learning and Residual CNNs
- 30 October 2025
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 113
PBSNet: Parallel branch spatial network for precise brain tumor classification in MRI slices
- 30 October 2025
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 113
PBSNet: Parallel branch spatial network for precise brain tumor classification in MRI slices
- Open Access
- 30 September 2025
- Published by The University of Danang in The University of Danang - Journal of Science and Technology
- p. 80-85
Efficient chatbot for university admission consultation using large language models
This paper presents an Artificial Intelligence–driven chatbot for university admission consultation using Large Language Models (LLMs). The system integrates semantic retrieval with Retrieval-Augmented Generation (RAG) and employs a hybrid strategy that combines vector similarity and keyword matching to provide accurate and context-aware answers. The chatbot was trained on admission FAQs, official documents, and consultation records from Dong A University, ensuring relevance to real user needs. Implementation leverages efficient prompt construction and memory management to support interactive and personalized responses. Experimental results show improved retrieval precision and practical benefits in reducing staff workload and offering consistent support to prospective students. Current limitations include the use of a single-university dataset and a technical evaluation focused on retrieval metrics. Future work will expand to multi-institution data, user studies, and multilingual or voice-enabled interaction to enhance generalizability and real-world impact.
- 30 September 2025
- Published by The University of Danang in The University of Danang - Journal of Science and Technology
- p. 80-85
Open AccessEfficient chatbot for university admission consultation using large language models
This paper presents an Artificial Intelligence–driven chatbot for university admission consultation using Large Language Models (LLMs). The system integrates semantic retrieval with Retrieval-Augmented Generation (RAG) and employs a hybrid strategy that combines vector similarity and keyword matching to provide accurate and context-aware answers. The chatbot was trained on admission FAQs, official documents, and consultation records from Dong A University, ensuring relevance to real user needs. Implementation leverages efficient prompt construction and memory management to support interactive and personalized responses. Experimental results show improved retrieval precision and practical benefits in reducing staff workload and offering consistent support to prospective students. Current limitations include the use of a single-university dataset and a technical evaluation focused on retrieval metrics. Future work will expand to multi-institution data, user studies, and multilingual or voice-enabled interaction to enhance generalizability and real-world impact.
- Open Access
A Non-Monotonic Relationship: An Empirical Analysis of Hybrid Quantum Classifiers for Unseen Ransomware Detection
Detecting unseen ransomware is a critical cybersecurity challenge where classical machine learning often fails. While Quantum Machine Learning (QML) presents a potential alternative, its application is hindered by the dimensionality gap between classical data and quantum hardware. This paper empirically investigates a hybrid framework using a Variational Quantum Classifier (VQC) interfaced with a high-dimensional dataset via Principal Component Analysis (PCA). Our analysis reveals a dual challenge for practical QML. A significant information bottleneck was evident, as even the best performing 12-qubit VQC fell short of the classical baselines 97.7\% recall. Furthermore, a non-monotonic performance trend, where performance degraded when scaling from 4 to 8 qubits before improving at 12 qubits suggests a severe trainability issue. These findings highlight that unlocking QMLs potential requires co-developing more efficient data compression techniques and robust quantum optimization strategies.
Open AccessA Non-Monotonic Relationship: An Empirical Analysis of Hybrid Quantum Classifiers for Unseen Ransomware Detection
Detecting unseen ransomware is a critical cybersecurity challenge where classical machine learning often fails. While Quantum Machine Learning (QML) presents a potential alternative, its application is hindered by the dimensionality gap between classical data and quantum hardware. This paper empirically investigates a hybrid framework using a Variational Quantum Classifier (VQC) interfaced with a high-dimensional dataset via Principal Component Analysis (PCA). Our analysis reveals a dual challenge for practical QML. A significant information bottleneck was evident, as even the best performing 12-qubit VQC fell short of the classical baselines 97.7\% recall. Furthermore, a non-monotonic performance trend, where performance degraded when scaling from 4 to 8 qubits before improving at 12 qubits suggests a severe trainability issue. These findings highlight that unlocking QMLs potential requires co-developing more efficient data compression techniques and robust quantum optimization strategies.