Publications

2026

1. CSoNet

   

   A Federated Benchmark for Multimodal Emotion Recognition under Speaker-Level Data Heterogeneity

   Thu-Ky Nguyen ^†, Nhat Khanh Le ^†, Nhut Minh Nguyen ^† , Ngoc-Khoa Le, and Duc Ngoc Minh Dang ^*

   In Proceedings of The 15th International Conference on Computational Science and Network Intelligence, Nov 2026

   To appear.

   Abs Bib Code

   Multimodal Emotion Recognition (MER) has achieved significant progress under centralized training settings. However, centralized training poses substantial privacy risks because speech and textual data are inherently sensitive. Federated Learning enables collaborative training without sharing raw data, yet the behavior of different MER architectures under Federated Learning remains largely underexplored. This paper presents FedMER, a systematic benchmark of four MER architectures and five Federated Learning strategies under a unified speaker-shard non-independent and identically distributed (non-IID) protocol. We conduct experiments on two benchmark datasets, providing reproducible and comparable evaluations under realistic privacy-preserving conditions. The results demonstrate that optimization and aggregation techniques alone do not consistently mitigate the effects of speaker-level non-IID data. Federated Averaging (FedAvg) provides a strong and stable baseline, while Federated Normalized Averaging (FedNova) achieves comparable performance by correcting for unequal local training progress. In contrast, the proximal regularization used by Federated Proximal Optimization (FedProx) consistently degrades performance across both datasets. Techniques based on local normalization and prototype-guided aggregation provide benefits only under specific model-dataset combinations and do not universally outperform standard parameter averaging. Overall, fusion design remains an important determinant of recognition performance across the evaluated federated settings, while the remaining gap from centralized training indicates that speaker-level heterogeneity is not fully addressed by existing federated optimization techniques.

   @inproceedings{Nguyen2026FedMER,
     bibtex_show = true,
     title = {A Federated Benchmark for Multimodal Emotion Recognition under Speaker-Level Data Heterogeneity},
     author = {Nguyen , Thu-Ky and Le , Nhat Khanh and Nguyen , Nhut Minh and Le, Ngoc-Khoa and Dang , Duc Ngoc Minh},
     year = {2026},
     month = nov,
     address = {Ho Chi Minh City, Vietnam},
     booktitle = {Proceedings of The 15th International Conference on Computational Science and Network Intelligence},
     note = {To appear.}
   }

2. FISAT

   

   Multimodal Disentangled Network with Alignment-Aggregation Decoupled Asymmetric Cross-Attention for Emotion Recognition

   Quoc Anh Phan , Nhut Minh Nguyen, and Duc Ngoc Minh Dang

   In Proceedings of 1st EAI FPT International Conference on Intelligent Systems and Advanced Technologies, Nov 2026

   To appear.

   Abs Bib

   Multimodal emotion recognition (MER) with speech and text requires fusion that aligns modalities while preserving modality-specific emotional cues. We propose a Multimodal Disentangled Network with Alignment-Aggregation Decoupled Asymmetric Cross-Attention. It separates each modality into Private, Auxiliary, and Shared subspaces. Private branches retain fine-grained cues, Auxiliary branches learn Central Moment Discrepancy (CMD)-aligned task-relevant features, and the Shared branch captures modality-invariant patterns through the InfoNCE contrastive loss. Query and Key are computed from Auxiliary representations, whereas Value is drawn from Private representations, decoupling cross-modal correspondence estimation from content aggregation. Experiments on the Interactive Emotional Dyadic Motion Capture (IEMOCAP) and Multimodal EmotionLines Dataset (MELD) benchmarks show competitive results. On four-class IEMOCAP with five-fold Leave-One-Session-Out (LOSO), the model achieves 77.82% weighted accuracy (WA) and 79.03% unweighted accuracy (UA), best among compared methods. On seven-class MELD, it obtains 64.65% WA and 61.22% weighted F1 (WF1), outperforming most baselines and remaining competitive with state-of-the-art results. Ablations indicate that Auxiliary branches, asymmetric attention routing, and branch-specific supervision are key contributors. These results support decoupled alignment and aggregation as an effective principle for audio-text emotion recognition.

   @inproceedings{Phan2026FISAT,
     bibtex_show = true,
     title = {Multimodal Disentangled Network with Alignment-Aggregation Decoupled Asymmetric Cross-Attention for Emotion Recognition},
     author = {Phan, Quoc Anh and Nguyen, Nhut Minh and Dang, Duc Ngoc Minh},
     year = {2026},
     month = nov,
     address = {Ho Chi Minh City, Vietnam},
     booktitle = {Proceedings of 1st EAI FPT International Conference on Intelligent Systems and Advanced Technologies},
     note = {To appear.}
   }

3. APSIPA ASC

   

   Prompt-Guided Missing Modality Completion for Multimodal Emotion Recognition

   Nhut Minh Nguyen, Nhat Truong Pham, Quang Nhan Hoang, Hanh Dang-Ngoc , Truc Mai Anh Nguyen, and Duc Ngoc Minh Dang

   In Proceedings of 2026 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), Nov 2026

   To appear.

   Abs Bib Code

   Multimodal Emotion Recognition (MER) is an effective approach to recognizing human emotions by combining complementary information from multiple modalities. However, in real-world scenarios, collecting complete multimodal data during inference is challenging due to environmental noise and errors in automatic speech recognition. These issues lead to missing modality situations, degrading the performance of MER systems. To address these challenges, we propose Prompt4MER, a prompt-learning-based framework designed to enhance MER robustness under missing modality conditions. Prompt4MER employs a two-stage training strategy, including pre-training with all modalities and fine-tuning under missing modality conditions. The model bridges the gap between training with complete data and inference with incomplete data. We further introduce a Learnable Missing Modality Prompt (LMMP) module that combines modality-specific prompts and missing modality prompts to provide learnable missing-modality surrogate tokens that maintain modality-aware presentation under incomplete inputs. In addition, we design a Text-Guided Cross-Modal Fusion module that uses textual information as a semantic anchor to integrate complementary audio and visual cues. Experiments on IEMOCAP and MSP-IMPROV under six missing modality configurations demonstrate that Prompt4MER achieves strong average performance across six missing-modality configurations compared with existing missing-modality MER baselines.

   @inproceedings{Nguyen2026Prompt4MER,
     bibtex_show = true,
     title = {Prompt-Guided Missing Modality Completion for Multimodal Emotion Recognition},
     author = {Nguyen, Nhut Minh and Pham, Nhat Truong and Hoang, Quang Nhan and Dang-Ngoc, Hanh and Nguyen, Truc Mai Anh and Dang, Duc Ngoc Minh},
     year = {2026},
     month = nov,
     address = {Bangkok, Thailand},
     booktitle = {Proceedings of 2026 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC)},
     note = {To appear.}
   }

4. APSIPA ASC

   

   Relational and Sequential Learning for Transition-Aware Multimodal Emotion Recognition in Conversation

   Quoc Bao Nguyen, Hoang Khang Trang , Nhut Minh Nguyen, and Duc Ngoc Minh Dang

   In Proceedings of 2026 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), Nov 2026

   To appear.

   Abs Bib Code

   Multimodal Emotion Recognition in Conversation (ERC) aims to effectively integrate textual, acoustic, and visual cues to capture human interactions in intelligent service systems. However, existing methods often model conversational context in a unified manner without explicitly capturing how emotions evolve across consecutive utterances. To address these limitations, we propose RSL-TAM, a robust framework that integrates speaker-aware relational subgraphs with reinforcement learning (RL) for Multimodal ERC. Specifically, we construct two relational subgraphs to model emotional continuity within the same speaker and emotional influences across different speakers. The proposed design allows the model to capture speaker-specific and cross-speaker dependencies through separate Graph Attention Network (GAT) based learning. Then, the RL module treats the fused multimodal hidden representation as sequential states to feed into a Q-value policy head and optimizes the policy head with a Bellman-based temporal-difference objective and an emotion-shift-aware reward. This encourages the final fused representations to become more consistent and better capture emotion transitions across consecutive utterances. Experimental results demonstrate that RSL-TAM outperforms current competitive methods by 3.79% in Accuracy on the IEMOCAP dataset and 0.58% in Accuracy on the MELD dataset.

   @inproceedings{Nguyen2026RSL-TAM,
     bibtex_show = true,
     title = {Relational and Sequential Learning for Transition-Aware Multimodal Emotion Recognition in Conversation},
     author = {Nguyen, Quoc Bao and Trang, Hoang Khang and Nguyen, Nhut Minh and Dang, Duc Ngoc Minh},
     year = {2026},
     month = nov,
     address = {Bangkok, Thailand},
     booktitle = {Proceedings of 2026 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC)},
     note = {To appear.}
   }

5. ATC

   

   Audio-Guided Repair Gate for ASR-Robust Speech Emotion Recognitionn

   Nhut Minh Nguyen, Duc Tai Phan , Duc Nguyen Huu Dang , Thu Ky Nguyen, and Duc Ngoc Minh Dang

   In Proceedings of 2026 International Conference on Advanced Technologies for Communications (ATC), Oct 2026

   To appear.

   Abs Bib Code

   Speech Emotion Recognition (SER) is an important task in affective computing and human-computer interaction. While multimodal SER can leverage complementary audio and textual cues, real-world systems often rely on Automatic Speech Recognition (ASR)-generated transcripts instead of manual transcripts. However, ASR errors may degrade semantic representations and reduce emotion recognition performance. In this study, we propose AURORA, an audio-guided repair gate architecture for ASR-robust SER. AURORA adopts a teacher-student architecture, where the teacher branch learns from audio and clean text, while the student branch learns from audio and ASR-generated text to match real-world inference conditions. The student branch is further equipped with an audio-guided repair gate, which pairs a repair residual with a reliability gate to selectively correct ASR-based textual representations. Gated multimodal fusion and knowledge distillation are used to integrate multimodal information and transfer reliable knowledge from the teacher branch to the student branch. We evaluate performance using Weighted Accuracy (WA), which measures overall classification accuracy across all samples, and Unweighted Accuracy (UA), which averages class-wise accuracy to account for class imbalance. Experiments on two benchmark datasets show that AURORA achieves competitive performance, with 79.96% WA and 79.77% UA on IEMOCAP, and 63.27% WA and 62.47% UA on MSP-IMPROV. Ablation studies validate the contribution of the proposed components to ASR-robust emotion recognition.

   @inproceedings{Nguyen2026AURORA,
     bibtex_show = true,
     title = {Audio-Guided Repair Gate for ASR-Robust Speech Emotion Recognitionn},
     author = {Nguyen, Nhut Minh and Phan, Duc Tai and Dang, Duc Nguyen Huu and Nguyen, Thu Ky and Dang, Duc Ngoc Minh},
     year = {2026},
     month = oct,
     address = {Quy Nhon City, Gia Lai Province, Vietnam},
     booktitle = {Proceedings of 2026 International Conference on Advanced Technologies for Communications (ATC)},
     note = {To appear.}
   }

6. ATC

   

   A Domain-Adaptive Kolmogorov-Arnold Network for Cross-Subject EEG-Based Emotion Recognition

   Duc Nguyen Huu Dang , Nhut Minh Nguyen, Quang Nhan Hoang , Nhat Khanh Le, and Duc Ngoc Minh Dang

   In Proceedings of 2026 International Conference on Advanced Technologies for Communications (ATC), Oct 2026

   To appear.

   Abs Bib Code

   Electroencephalogram (EEG)-based emotion recognition encounters considerable challenges due to inter-subject variability and complex spatio-temporal dependencies, limiting the generalization of existing models in cross-subject contexts. This work presents DoGKAN, a hybrid architecture capable of integrating spatial dependency modeling, temporal sequence learning, and cross-subject domain alignment within a unified framework. Specifically, a Graph Convolutional Network (GCN) is employed to capture spatial interconnections among EEG electrodes, while a Transformer encoder models temporal dynamics of EEG sequences. To enhance nonlinear representation capacity, we replace traditional fully connected layers with Kolmogorov-Arnold Network (KAN), which utilizes learnable spline-based nonlinear mappings to provide greater representational flexibility and built-in sparsity regularization. Furthermore, a Domain-Adversarial Neural Network (DANN) with a Gradient Reversal Layer (GRL) is incorporated to mitigate cross-subject distribution discrepancies. The model’s average accuracy under the Leave-One-Subject-Out (LOSO) protocol reaches 77.89% on SEED-IV and 84.86% on SEED-V. The results highlight the effectiveness of applying graph-based spatial modeling, Transformer-based temporal learning, and KAN-driven adaptive nonlinear classification for robust cross-subject EEG-based affective computing.

   @inproceedings{Dang2026DoGKAN,
     bibtex_show = true,
     title = {A Domain-Adaptive Kolmogorov-Arnold Network for Cross-Subject EEG-Based Emotion Recognition},
     author = {Dang, Duc Nguyen Huu and Nguyen, Nhut Minh and Hoang, Quang Nhan and Le, Nhat Khanh and Dang, Duc Ngoc Minh},
     year = {2026},
     month = oct,
     address = {Quy Nhon City, Gia Lai Provide, Vietnam},
     booktitle = {Proceedings of 2026 International Conference on Advanced Technologies for Communications (ATC)},
     note = {To appear.}
   }

7. ICIT

   

   A UAV-Enabled Framework for Early Rip-Current Warning Using Lightweight Patch-Based Inference and Temporally Stabilized Risk Estimation

   Hanh Dang-Ngoc , Nhut Minh Nguyen, and Duc Ngoc Minh Dang

   In Proceedings of the 5th International Conference on Intelligence of Things 2026, Oct 2026

   To appear.

   Abs Bib

   Rip currents are major hazards at surf beaches, and timely warning requires identifying high-risk nearshore regions before swimmers are exposed to dangerous offshore flows. This paper proposes an unmanned aerial vehicle (UAV)-enabled framework in which aerial red-green-blue (RGB) video is streamed to a ground station for near-real-time rip-current warning. The framework targets practical deployment conditions where inference must remain lightweight, interpretable, and robust to variations in viewpoint, sea state, illumination, wave breaking, and shoreline appearance. The proposed pipeline tiles each frame into overlapping patches, extracts features using a frozen MobileNetV3-Large backbone, and applies a lightweight classifier to estimate rip-current likelihood. Patch scores are fused into a spatial score map and temporally smoothed using exponential moving average (EMA) to produce a stabilized operator-facing risk estimate over the image plane. RipAID is used for training, validation, and in-distribution testing, while RipVIS-val is used for no-tuning external evaluation. RipVIS polygon and mask annotations are converted to oriented bounding boxes (OBBs) for consistent processing. Experimental results show that the linear head achieves a precision of 0.899, recall of 0.985, and 𝐹2 score of 0.966 on RipAID-test. On RipVIS-val, EMA improves 𝐹2 from 0.870 to 0.889 and reduces Δ𝑀 from 0.245 to 0.021. The results indicate that lightweight UAV-assisted warning can produce temporally stabilized risk estimates suitable for operator review.

   @inproceedings{Dang2026UAV,
     bibtex_show = true,
     title = {A UAV-Enabled Framework for Early Rip-Current Warning Using Lightweight Patch-Based Inference and Temporally Stabilized Risk Estimation},
     author = {Dang-Ngoc, Hanh and Nguyen, Nhut Minh and Dang, Duc Ngoc Minh},
     year = {2026},
     month = oct,
     address = {Sejong University, Seoul, Republic of Korea},
     booktitle = {Proceedings of the 5th International Conference on Intelligence of Things 2026},
     note = {To appear.}
   }

8. APWeb

   

   CorrTie: Correction-aware Tie-breaking for Active Learning with Vision-Language Models

   Duc Tai Phan, Tri Nguyen Thai , Khoa Minh Nguyen , Nhut Minh Nguyen, and Duc Ngoc Minh Dang

   In Proceedings of the 10th APWeb-WAIM joint International Conference on Web and Big Data, Sep 2026

   Abs Bib Publication Code

   Vision-language models (VLMs) provide strong transferable representations for visual recognition, but adapting them to downstream image classification tasks remains challenging when labeled data are scarce. Active learning (AL) reduces annotation cost by selecting informative unlabeled samples for oracle labeling; however, most AL strategies use oracle feedback only to expand the labeled training set. The correction event between a model prediction and the oracle label is usually not reused to guide future acquisition. To address this limitation, we propose Correction-aware Tie-breaking (CorrTie), a lightweight post-acquisition reranker for AL with VLMs. CorrTie keeps the base AL strategy unchanged and uses past oracle corrections to estimate class-wise correction risk. This risk is then used to rerank near-tie candidates while preserving the original acquisition ranking as the primary signal. Experiments with CoOp prompt learning on four image classification benchmarks and five base AL strategies show that CorrTie improves the average accuracy by +7.14, +3.74, and +0.66 percentage points at the 1%, 2%, and 5% annotation budgets, respectively. These results indicate that oracle corrections can serve not only as supervision for prompt learning, but also as useful feedback for future active sample acquisition.

   @inproceedings{phan2026CorrTie,
     bibtex_show = true,
     title = {CorrTie: Correction-aware Tie-breaking for Active Learning with Vision-Language Models},
     author = {Phan, Duc Tai and Thai, Tri Nguyen and Nguyen, Khoa Minh and Nguyen, Nhut Minh and Dang, Duc Ngoc Minh},
     year = {2026},
     month = sep,
     address = {Da Nang City, Vietnam},
     booktitle = {Proceedings of the 10th APWeb-WAIM joint International Conference on Web and Big Data},
     doi = {10.1007/978-981-92-5699-0_26},
   }

9. INS

   

   Towards effective multimodal emotion recognition in conversations via joint multi-graph learning and self-distillation

   Hoang Khang Trang , Bao Quoc Nguyen, Viet Vinh Khanh Dinh , Nhut Minh Nguyen, Nhat Truong Pham, Phuong-Nam Tran, Phuong Luu Vo, and Duc Ngoc Minh Dang

   Information Sciences, Sep 2026

   Abs Bib Publication Code

   Emotion Recognition in Conversation (ERC) is a crucial component of empathetic dialogue systems, enabling conversational agents to comprehend users’ expressed emotions and respond appropriately. In recent years, ERC has increasingly evolved toward multimodal approaches, particularly graph-based ones, that leverage complementary signals to enhance effectiveness. Although these approaches show promise, they face challenges in effectively modeling relational structures, computing the contribution of each modality during fusion, and regularizing unimodal representations to prevent modality bias. In this study, we propose DiGemo, a novel multimodal ERC framework that employs Transformer-based contextual modeling and graph-based learning to jointly capture both sequential dependencies within conversations and complex inter-utterance relational structures across modalities. To effectively balance and exploit multimodal information, we employ an adaptive fusion approach based on the Gated Multimodal Unit to estimate the contributions of each modality. In addition, we integrate a self-distillation training strategy to regularize unimodal representations, allowing each modality to learn from the fused multimodal representation and achieve more stable learning. DiGemo achieves Weighted F1-scores of 74.44% and 65.92% on the IEMOCAP and MELD datasets, respectively. These results validate the effectiveness of jointly modeling conversational context, cross-modal interactions, adaptive contributions of modalities, and self-distillation for multimodal ERC.

   @article{TRANG2026124138,
     title = {Towards effective multimodal emotion recognition in conversations via joint multi-graph learning and self-distillation},
     journal = {Information Sciences},
     pages = {124138},
     year = {2026},
     issn = {0020-0255},
     doi = {10.1016/j.ins.2026.124138},
     author = {Trang, Hoang Khang and Nguyen, Bao Quoc and Dinh, Viet Vinh Khanh and Nguyen, Nhut Minh and Pham, Nhat Truong and Tran, Phuong-Nam and Vo, Phuong Luu and Dang, Duc Ngoc Minh},
     keywords = {Gated multimodal unit, Graph learning, Self-distillation, Emotion recognition in conversation},
     bibtex_show = true,
   }

10. TJIT

    

    Enhancing handwritten mathematical expression recognition with convolutional block attention refinement module and multi-task learning

    Thanh Trung Nguyen , Nhut Minh Nguyen, Nam Van Hai Phan , Khoa Minh Nguyen , Trung Thanh Pham, Phuong-Nam Tran , Cuong Tuan Nguyen, and Duc Ngoc Minh Dang

    Journal of Information and Telecommunication, Sep 2026

    Abs Bib Publication Code

    Handwritten Mathematical Expression Recognition (HMER) is challenging due to complex structural relationships, implicit symbols, and variations in handwriting styles. Therefore, this paper presents a novel approach combining multi-task learning (MTL) with advanced attention mechanisms, including the Convolutional Block Attention Module (CBAM) integrated into the Attention Refinement Module (ARM), to optimize attention mechanisms, enable context-aware interpretation, and improve the generalization of the architecture. Integrating MTL enhances the model’s ability to work well on the main task. By sharing knowledge across tasks, the model reduces overfitting and leads to better generalization in the inference stage. Advanced attention mechanisms refine the attention weights to effectively capture highly complex structural expressions and implicit corrections. Our approach addresses the limitations of traditional methods and improves the Expression Recognition Rate on CROHME datasets. Experimental results show that our model has competitive performance and fewer parameters when compared with the state-of-the-art.

    @article{Nguyen19082026,
      author = {Nguyen, Thanh Trung and Nguyen, Nhut Minh and Phan, Nam Van Hai and Nguyen, Khoa Minh and Pham, Trung Thanh and Tran, Phuong-Nam and Nguyen, Cuong Tuan and Dang, Duc Ngoc Minh},
      bibtex_show = true,
      title = {Enhancing handwritten mathematical expression recognition with convolutional block attention refinement module and multi-task learning},
      journal = {Journal of Information and Telecommunication},
      volume = {0},
      number = {0},
      pages = {1--33},
      year = {2026},
      publisher = {Taylor \& Francis},
      doi = {10.1080/24751839.2026.2716432},
      eprint = {https://doi.org/10.1080/24751839.2026.2716432},
    }

11. ECTI-CON

    

    Swin Transformer V2 for Optical Chemical Structure Recognition: Comparison with Convolutional Neural Networks and Swin Transformer Variants

    Thanh Trung Nguyen , Nhut Minh Nguyen, Duc Tai Phan, Quang Nhan Hoang , Tri Minh Pham, and Duc Ngoc Minh Dang

    In The 23rd International Conference on Electrical Engineering/Electronics, Computer, Telecommunication, and Information Technology, Sep 2026

    Abs Bib Publication

    Optical Chemical Structure Recognition (OCSR) is designed to transform images of chemical structures into formats that computers can process, facilitating the extraction of molecular data from scientific publications on a large scale. Although modern deep learning methods have shown encouraging results, the impact of different visual backbone designs on understanding molecular structures has not been thoroughly investigated. This research compares various visual backbones for OCSR using an encoder-decoder architecture. The results show that while Convolutional Neural Network (CNN) architectures deliver excellent accuracy at the character level, Swin Transformer V2 generates more accurate molecular structures, as evidenced by superior Tanimoto similarity metrics. Moreover, even though Swin Transformer V2 has similar computational demands as the original Swin Transformer (V1), it exhibits enhanced training reliability and better capability for structural modeling, underscoring its effectiveness as a dependable visual backbone for OCSR applications.

    @inproceedings{Nguyen2025Ecti-Con,
      bibtex_show = true,
      author = {Nguyen, Thanh Trung and Nguyen, Nhut Minh and Phan, Duc Tai and Hoang, Quang Nhan and Pham, Tri Minh and Dang, Duc Ngoc Minh},
      title = {Swin Transformer V2 for Optical Chemical Structure Recognition: Comparison with Convolutional Neural Networks and Swin Transformer Variants},
      booktitle = {The 23rd International Conference on Electrical Engineering/Electronics, Computer, Telecommunication, and Information Technology},
      address = {Chonburi, Thailand},
      volume = {1},
      pages = {299--304},
      year = {2026},
      doi = {10.1109/ECTI-CON68836.2026.11608590},
    }

12. KBS

    

    From object difficulty to image scoring: A strategy for active learning in object detection

    Duc Tai Phan , Nhut Minh Nguyen , Khang Phuc Nguyen, Phuong-Nam Tran, Nhat Truong Pham, Linh Le, Choong Seon Hong, and Duc Ngoc Minh Dang

    Knowledge-Based Systems, Apr 2026

    Abs Bib Publication Code

    Object detection (OD) faces the costly hurdle of large-scale dataset annotation. Active Learning (AL) is a promising solution that selects the most beneficial samples for annotation. Applying AL to OD is challenging, as it requires addressing uncertainties in both classifying and localizing objects while combining object-level information into a single image-level decision. Current methods, like uncertainty sampling, often rely on output-level uncertainty or heuristic signal combinations. However, relying on static model outputs limits their ability to capture underlying feature-space instabilities. To address these limitations, we propose Feature Difficulty-based AL (FDAL), a novel framework that shifts the focus from output-level heuristics to active latent-space manipulation. By systematically interpolating unlabeled features toward class anchors, FDAL identifies prediction inconsistencies to reveal hidden traits unknown to the model. This anchor-guided interpolation unifies classification and localization uncertainties, capturing latent instabilities that traditional paradigms miss. Experiments on four OD benchmarks demonstrate that FDAL achieves state-of-the-art (SOTA) results, consistently improving detection accuracy while substantially reducing annotation costs. FDAL outperforms SOTA methods by 0.8% in detection accuracy on Pattern analysis, statistical modelling, and computational learning (PASCAL) Visual Object Classes (PASCAL VOC), 1.99% on Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI), 1.68% on Cityscapes, and 0.27% on Microsoft Common Objects in Context (MS COCO). FDAL’s selection is highly efficient, taking 0.06 seconds per image on PASCAL VOC and up to 0.63 seconds per image on Cityscapes, enabling scalability to large unlabeled pools. FDAL offers a practical, efficient, and effective solution for advancing AL in OD.

    @article{PHAN2026115946,
      author = {Phan, Duc Tai and Nguyen, Nhut Minh and Nguyen, Khang Phuc and Tran, Phuong-Nam and Pham, Nhat Truong and Le, Linh and Hong, Choong Seon and Dang, Duc Ngoc Minh},
      title = {From object difficulty to image scoring: A strategy for active learning in object detection},
      journal = {Knowledge-Based Systems},
      month = apr,
      year = {2026},
      keywords = {Active learning, Object detection, Computer vision, Difficulty, Instability},
      volume = {342},
      pages = {115946},
      issn = {0950-7051},
      doi = {10.1016/j.knosys.2026.115946},
      url = {https://www.sciencedirect.com/science/article/pii/S0950705126006726},
    }

13. INIS

    

    HyperDyG: Hypergraph-Driven Dynamic Fusion for Semi-Supervised Multimodal Emotion Recognition

    Nhut Minh Nguyen , Thu Thuy Le , Thanh Trung Nguyen , Hanh Ngoc Dang, Luu Phuong Vo, and Duc Ngoc Minh Dang

    EAI Endorsed Transactions on Industrial Networks and Intelligent Systems Journal, Feb 2026

    Abs Bib Publication Code

    Speech emotion recognition (SER) is important in healthcare, education, human–computer interaction, and customer service. Multimodal emotion recognition (MER) integrates audio and textual modalities to achieve a comprehensive understanding of human affect, but still suffers from limited labeled data and complex cross-modal relations.  To address these challenges, we propose HyperDyG, a dynamic hypergraph-driven MER framework. The HyperDyG leverages the strengths of dynamic hypergraph learning (DHL), cross-modal transformer (CMT), and an adaptive gated multimodal unit (GMU) for robust multimodal fusion. HyperDyG is further enhanced with a semi-supervised learning strategy that incorporates weak–strong augmentation, confidence-filtered pseudo-labeling, and consistency regularization to effectively exploit large-scale unlabeled data. The HyperDyG achieves state-of-the-art (SOTA) performance on the benchmark emotion dataset and maintains stable accuracy across varying unlabeled ratios. The findings of HyperDyG highlight the effectiveness and scalability of the proposed architecture in real-world low-label MER scenarios.

    @article{nguyen2026HyperDyG,
      bibtex_show = true,
      title = {HyperDyG: Hypergraph-Driven Dynamic Fusion for Semi-Supervised Multimodal Emotion Recognition},
      author = {Nguyen, Nhut Minh and Le, Thu Thuy and Nguyen, Thanh Trung and Dang, Hanh Ngoc and Vo, Luu Phuong and Dang, Duc Ngoc Minh},
      journal = {EAI Endorsed Transactions on Industrial Networks and Intelligent Systems Journal},
      volume = {13},
      year = {2026},
      month = feb,
      doi = {10.4108/eetinis.131.10903},
    }

14. MLHMI

    

    SemiFedER: Semi-supervised Federated Averaging for Multimodal Emotion Recognition

    Nhut Minh Nguyen , Thu Thuy Le , Thanh Trung Nguyen, and Duc Ngoc Minh Dang

    In Proceedings of the 7th International Conference on Machine Learning and Human-Computer Interaction (MLHMI 2026), Mar 2026

    Abs Bib Publication Code

    Multimodal Emotion Recognition (MER) is a powerful approach for human–computer interaction, leveraging complementary cues across multiple modalities to infer human affect. MER has broad real-world potential in applications such as healthcare monitoring and affect-aware virtual assistants. However, practical deployments face two major challenges: large-scale data is often unlabeled due to the high cost of emotion annotation, and speech signals and transcripts are privacy-sensitive, limiting centralized data collection and training. To address these limitations, we propose SemiFedER, a semi-supervised federated learning framework for MER. SemiFedER performs client-side training in two stages, including supervised pre-training on labeled samples and semi-supervised learning that exploits unlabeled data via confidence-based pseudo-labeling and weak-strong consistency regularization. The server aggregates client updates using Federated Averaging (FedAvg) to learn a global model without sharing raw data. We deploy the representative centralized MER backbones within SemiFedER to assess their effectiveness in this practical setting. Extensive experiments on the MELD dataset under speaker-disjoint non-IID federated splits demonstrate that SemiFedER provides stable performance across labeled ratios and client counts, and achieves competitive improvements in class-balanced evaluation compared to centralized baselines.

    @inproceedings{Nguyen2026SemiFedER,
      bibtex_show = true,
      title = {SemiFedER: Semi-supervised Federated Averaging for Multimodal Emotion Recognition},
      author = {Nguyen, Nhut Minh and Le, Thu Thuy and Nguyen, Thanh Trung and Dang, Duc Ngoc Minh},
      booktitle = {Proceedings of the 7th International Conference on Machine Learning and Human-Computer Interaction (MLHMI 2026)},
      year = {2026},
      address = {Tokyo, Japan},
      month = mar,
      doi = {10.23919/MLHMICPS00004.2026.00038},
    }

15. AEEE

    

    HemoGAT: Heterogeneous Multimodal Speech Emotion Recognition with Cross-Modal Transformer and Graph Attention Network

    Nhut Minh Nguyen , Thanh Trung Nguyen , Tien-Dat Nguyen, and Duc Ngoc Minh Dang

    Advances in Electrical and Electronic Engineering, Jun 2026

    Abs Bib Publication Code

    Multimodal speech emotion recognition (SER) is a promising field, yet effectively fusing diverse information streams remains challenging. Addressing this requires architectures capable of modeling structural relationships across modalities with fine-grained, feature-level interactions. This paper proposes HemoGAT, a novel heterogeneous multimodal SER architecture that integrates a dual-stream architecture with two core modules: a heterogeneous multimodal graph attention network (HM-GAT) and a cross-modal transformer (CMT) to address this. The HM-GAT module captures complex structural and contextual dependencies using a heterogeneous graph constructed from deep embeddings. The CMT module enables precise cross-modal feature fusion through bidirectional cross-attention. This design effectively captures both high-level relationships and immediate cross-modal influences. HemoGAT achieves state-of-the-art (SOTA) performance on the IEMOCAP dataset and highly competitive results on the MELD dataset, demonstrating its superiority over existing methods. Extensive ablation studies were conducted to evaluate HemoGAT. We assessed the impact of the Top-K algorithm for heterogeneous graph construction and compared unimodal and multimodal fusion strategies. We also examined the contributions of the HM-GAT and CMT modules, analyzed the role of the graph attention network (GAT) in graph learning, and evaluated the effect of GAT layer depth on performance.

    @article{nguyen2026HemoGAT,
      bibtex_show = true,
      title = {HemoGAT: Heterogeneous Multimodal Speech Emotion Recognition with Cross-Modal Transformer and Graph Attention Network},
      author = {Nguyen, Nhut Minh and Nguyen, Thanh Trung and Nguyen, Tien-Dat and Dang, Duc Ngoc Minh},
      journal = {Advances in Electrical and Electronic Engineering},
      year = {2026},
      month = jun,
      doi = {10.15598/aeee.v24i2.250415},
    }

16. EAAI

    

    Enhancing multimodal emotion recognition with dynamic fuzzy membership and attention fusion

    Nhut Minh Nguyen , Minh Trung Nguyen , Thanh Trung Nguyen, Phuong-Nam Tran, Nhat Truong Pham, Linh Le, Alice Othmani, Abdulmotaleb El Saddik, and Duc Ngoc Minh Dang

    Engineering Applications of Artificial Intelligence, Feb 2026

    Abs Bib Publication Code

    Multimodal learning has been shown to enhance classification outcomes in speech emotion recognition (SER). Despite this advantage, multimodal approaches in SER often face key challenges, including limited robustness to uncertainty, difficulty generalizing across diverse emotional contexts, and inefficiencies in integrating heterogeneous modalities. To overcome these constraints, we propose a multimodal emotion recognition architecture, named FleSER, which leverages dynamic fuzzy membership and attention-based fusion. Unlike most previous SER studies that apply fuzzy logic at the decision level, FleSER introduces a feature-level, rule-based dynamic fuzzy membership mechanism that adaptively refines modality representations prior to fusion. The FleSER architecture leverages audio and textual modalities, employing self-modality and cross-modality attention mechanisms with the α interpolation to capture complementary emotional cues. The α interpolation-based feature fusion mechanism adaptively emphasizes the more informative modality across varying contexts, ensuring robust multimodal integration. This comprehensive design enhances recognition accuracy. We evaluate FleSER on multiple benchmark datasets, surpassing previous state-of-the-art (SOTA) approaches and demonstrating superior effectiveness in emotion recognition. Ablation studies further validate the effectiveness of each key component, including unimodal and multimodal input effectiveness, fuzzy membership functions, fusion strategies, and the projection dimension, on the performance of the FleSER architecture.

    @article{nguyen2026fleser,
      title = {Enhancing multimodal emotion recognition with dynamic fuzzy membership and attention fusion},
      author = {Nguyen, Nhut Minh and Nguyen, Minh Trung and Nguyen, Thanh Trung and Tran, Phuong-Nam and Pham, Nhat Truong and Le, Linh and Othmani, Alice and Saddik, Abdulmotaleb El and Dang, Duc Ngoc Minh},
      journal = {Engineering Applications of Artificial Intelligence},
      year = {2026},
      month = feb,
      doi = {10.1016/j.engappai.2025.113396},
    }

17. EAAI

    

    Multimodal fusion in speech emotion recognition: A comprehensive review of methods and technologies

    Nhut Minh Nguyen , Thanh Trung Nguyen, Phuong-Nam Tran, Chee Peng Lim, Nhat Truong Pham, and Duc Ngoc Minh Dang

    Engineering Applications of Artificial Intelligence, Jan 2026

    Abs Bib Publication Code

    Speech emotion recognition (SER) plays a crucial role in human-computer interaction, enhancing numerous applications such as virtual assistants, healthcare monitoring, and customer support by identifying and interpreting emotions conveyed through spoken language. While single-modality SER systems demonstrate notable simplicity and computational efficiency, excelling in extracting critical features like vocal prosody and linguistic content, there is a pressing need to improve their performance in challenging conditions, such as noisy environments and the handling of ambiguous expressions or incomplete information. These challenges underscore the necessity of transitioning to multi-modal approaches, which integrate complementary data sources to achieve more robust and accurate emotion detection. With advancements in artificial intelligence, especially in neural networks and deep learning, many studies have employed advanced deep learning and feature fusion techniques to enhance SER performance. This review synthesizes comprehensive publications from 2020 to 2024, exploring prominent multi-modal fusion strategies, including early fusion, late fusion, deep fusion, and hybrid fusion methods, while also examining data representation, data translation, attention mechanisms, and graph-based fusion technologies. We assess the effectiveness of various fusion techniques across standard SER datasets, highlighting their performance in diverse tasks and addressing challenges related to data alignment, noise management, and computational demands. Additionally, we explore potential future directions for enhancing multi-modal SER systems, emphasizing scalability and adaptability in real-world applications. This survey aims to contribute to the advancement of multi-modal SER and to inform researchers about effective fusion strategies for developing more responsive and emotion-aware systems.

    @article{nguyen2025MSER,
      bibtex_show = true,
      title = {Multimodal fusion in speech emotion recognition: A comprehensive review of methods and technologies},
      author = {Nguyen, Nhut Minh and Nguyen, Thanh Trung and Tran, Phuong-Nam and Lim, Chee Peng and Pham, Nhat Truong and Dang, Duc Ngoc Minh},
      journal = {Engineering Applications of Artificial Intelligence},
      year = {2026},
      month = jan,
      doi = {10.1016/j.engappai.2025.112624},
    }

2025

1. BIGDATA

   

   Federated Semi-Supervised FixMatch: Enhancing CutMix for Medical Image Segmentation

   Thu Thuy Le , Nhut Minh Nguyen, Nhat Truong Pham, Phuong-Nam Tran, Nguyen Doan Hieu Nguyen, Phuong Luu Vo, Balachandran Manavalan, and Duc Ngoc Minh Dang

   In 2025 IEEE International Conference on Big Data, Jan 2025

   Abs Bib Publication

   Previous studies on federated learning (FL) often assume that each client has access to fully labeled data. In reality, however, most hospitals and clinical facilities lack sufficient labeled data, as annotation is costly and requires highly skilled experts. The key challenge, therefore, is how to effectively train FL models when only a small portion of data at each client is labeled. Federated semi-supervised learning (FSSL) has emerged as a promising solution. From the perspective of big data, this challenge is further intensified by the large scale, heterogeneity, and non-IID nature of data distributions. These factors significantly affect training efficiency, making it crucial to fully leverage diverse unlabeled datasets while ensuring privacy and minimizing communication costs for large-scale FL deployment. In this study, we apply data augmentation in FSSL to make better use of unlabeled data together with the limited amount of labeled data. Specifically, we investigate the role of CutMix, a widely used patch-level augmentation technique, when combined with Federated FixMatch (FedFixMatch). We evaluate different strategies, including applying CutMix only to labeled data, only to unlabeled data, to both branches, or to mixed data, and examine their effects under varying annotation conditions and heterogeneous client distributions. Additionally, we examine the impact of CutMix on communication efficiency, convergence dynamics, and training stability within FedFixMatch. To the best of our knowledge, this work provides the first comprehensive benchmark of CutMix strategies in FSSL for medical image segmentation. Our results provide important guidance for developing efficient augmentation strategies that address the practical challenges of big data and FL.

   @inproceedings{Le2025BigData,
     author = {Le, Thu Thuy and Nguyen, Nhut Minh and Pham, Nhat Truong and Tran, Phuong{-}Nam and Nguyen, Nguyen Doan Hieu and Vo, Phuong Luu and Manavalan, Balachandran and Dang, Duc Ngoc Minh},
     title = {Federated Semi-Supervised FixMatch: Enhancing CutMix for Medical Image Segmentation},
     booktitle = {2025 IEEE International Conference on Big Data},
     year = {2025},
     address = {Macau SAR, China},
     bibtex_show = true,
     doi = {10.1109/BigData66926.2025.11402514},
   }

2. RIVF

   

   YOLOv5-Powered Smart Parking System with IoT-Based Real-Time Slot Monitoring

   Phuong Lam Nguyen, Duc Tai Phan , Thanh Trung Nguyen , Nhut Minh Nguyen, and Duc Ngoc Minh Dang

   In Proceedings of the 2025 RIVF International Conference on Computing and Communication Technologies (RIVF), Jan 2025

   Abs Bib Publication

   Finding a parking space has become a daily struggle in an era of rapid urbanization and increasing car ownership, contributing to traffic congestion, lost time, and environmental stress. To transform urban parking management, this study presents a novel Smart Parking System (SPS) that seamlessly integrates deep learning and Internet of Things (IoT) technologies. Our SPS delivers a streamlined workflow, featuring real-time slot occupancy detection via a fine-tuned YOLOv5 model analyzing camera feeds, automated entry/exit handling triggered by IoT sensors, and a specialized License Plate Recognition (LPR) model. Additionally, it includes secure cloud-based data logging for effortless tracking and billing. The web dashboard provides real-time, color-coded visualizations of slot availability, supporting efficient parking management. Experimental evaluation demonstrates an accuracy of 90% in license plate recognition and over 99.5% in slot occupancy detection, highlighting the system’s reliability and practical applicability for smart city deployment.

   @inproceedings{Nguyen2025Parking,
     bibtex_show = true,
     author = {Nguyen, Phuong Lam and Phan, Duc Tai and Nguyen, Thanh Trung and Nguyen, Nhut Minh and Dang, Duc Ngoc Minh},
     booktitle = {Proceedings of the 2025 RIVF International Conference on Computing and Communication Technologies (RIVF)},
     title = {YOLOv5-Powered Smart Parking System with IoT-Based Real-Time Slot Monitoring},
     year = {2025},
     pages = {693-698},
     keywords = {YOLO;Analytical models;Smart cities;Automated parking;Real-time systems;Internet of Things;License plate recognition;Monitoring;Traffic congestion;Intelligent sensors;Smart Parking System;License Plate Recognition;Real-Time Slot Monitoring;YOLOv5;Optical Character Recognition;Computer Vision},
     doi = {10.1109/RIVF68649.2025.11365041},
   }

3. MEDES

   

   GloMER: Towards Robust Multimodal Emotion Recognition via Gated Fusion and Contrastive Learning

   Nhut Minh Nguyen, Duc Tai Phan, and Duc Ngoc Minh Dang

   In Proceedings of the 17th International Conference on Management of Digital Ecosystem (MEDES 2025), Nov 2025

   Abs Bib Publication Code

   Speech Emotion Recognition (SER) enhances Human-Computer Interaction (HCI) across healthcare, education, and customer service. Although multimodal approaches that combine audio and text show promise, they face challenges in aligning heterogeneous modalities and achieving balanced feature fusion. In this work, we propose GloMER, a novel multimodal SER architecture that integrates a self-alignment strategy with contrastive learning and a Gated Multimodal Unit (GMU) for adaptive fusion. The self-alignment mechanism ensures semantic consistency while preserving modality diversity. The GMU dynamically regulates modality contributions, ensuring balanced integration and context-aware representations under imbalanced conditions. We evaluate GloMER, achieving state-of-the-art (SOTA) on two benchmark datasets, delivering consistent improvements in multimodal emotion recognition. Ablation studies show that combining classification with alignment improves embedding discriminability, while gated fusion balances modalities and captures complementary cues. These findings establish GloMER as a robust framework for advancing multimodal emotion recognition.

   @inproceedings{nguyen2025GloMER,
     bibtex_show = true,
     title = {GloMER: Towards Robust Multimodal Emotion Recognition via Gated Fusion and Contrastive Learning},
     author = {Nguyen, Nhut Minh and Phan, Duc Tai and Dang, Duc Ngoc Minh},
     booktitle = {Proceedings of the 17th International Conference on Management of Digital Ecosystem (MEDES 2025)},
     year = {2025},
     month = nov,
     address = {Ho Chi Minh City, Vietnam},
     organization = {FPT University},
     doi = {10.1007/978-3-032-36547-7_7},
   }

4. MEDES

   

   DAAL: Dual Ambiguity in Active Learning for Object Detection with YOLOE

   Duc Tai Phan , Nhut Minh Nguyen, and Duc Ngoc Minh Dang

   In Proceedings of the 17th International Conference on Management of Digital Ecosystem (MEDES 2025), Nov 2025

   Abs Bib Publication Code

   The high cost of annotating vast datasets poses a significant challenge in object detection, hindering the development of robust models. By selectively annotating the most instructive samples, active learning has become a viable strategy for addressing this problem and improving model performance and labeling efficiency. However, traditional active learning techniques frequently fail to balance the two main types of model uncertainty: epistemic uncertainty, which captures uncertainty in object classification, and aleatoric uncertainty, which reflects inherent ambiguity in object presence and localization. In this paper, we introduce Dual Ambiguity Active Learning (DAAL), a novel framework that quantifies and combines both epistemic and aleatoric ambiguity into a single, weighted score. Epistemic ambiguity measures the model’s indecision in assigning semantic labels, while aleatoric ambiguity assesses the conviction in object presence and localization. By combining these, DAAL selects the most informative images for annotation, optimizing model performance under limited labeling budgets. Extensive experiments on popular benchmarks demonstrate that DAAL consistently outperforms traditional methods, achieving superior accuracy under the same limited labeling budget. This affirms its effectiveness in creating more efficient annotation workflows for object detection.

   @inproceedings{Phan2025DAAL,
     bibtex_show = true,
     title = {DAAL: Dual Ambiguity in Active Learning for Object Detection with YOLOE},
     author = {Phan, Duc Tai and Nguyen, Nhut Minh and Dang, Duc Ngoc Minh},
     booktitle = {Proceedings of the 17th International Conference on Management of Digital Ecosystem (MEDES 2025)},
     year = {2025},
     month = nov,
     address = {Ho Chi Minh City, Vietnam},
     organization = {FPT University},
     doi = {10.1007/978-3-032-36547-7_32},
   }

5. APNOMS

   

   CemoBAM: Advancing Multimodal Emotion Recognition through Heterogeneous Graph Networks and Cross-Modal Attention Mechanisms

   Nhut Minh Nguyen , Thu Thuy Le , Thanh Trung Nguyen, Duc Tai Phan , Anh Khoa Tran, and Duc Ngoc Minh Dang

   In Proceedings of the 2025 Asia-Pacific Network Operations and Management Symposium (APNOMS), Sep 2025

   Abs Bib Publication Code

   Multimodal Speech Emotion Recognition (SER) offers significant advantages over unimodal approaches by integrating diverse information streams such as audio and text. However, effectively fusing these heterogeneous modalities remains a significant challenge. We propose CemoBAM, a novel dualstream architecture that effectively integrates the Heterogeneous Graph Attention Network (CH-GAT) with the Cross-modal Convolutional Block Attention Mechanism (xCBAM). In CemoBAM architecture, the CH-GAT constructs a heterogeneous graph that models intra- and inter-modal relationships, employing multi-head attention to capture fine-grained dependencies across audio and text feature embeddings. The xCBAM enhances feature refinement through a cross-modal transformer with a modified 1D-CBAM, employing bidirectional cross-attention and channel-spatial attention to emphasize emotionally salient features. The CemoBAM architecture surpasses previous state-of-the-art (SOTA) methods by 0.32% on IEMOCAP and 3.25% on ESD datasets. Comprehensive ablation studies validate the impact of Top-K graph construction parameters, fusion strategies, and the complementary contributions of both modules. The results highlight CemoBAM’s robustness and potential for advancing multimodal SER applications.

   @inproceedings{nguyen2025CemoBAM,
     bibtex_show = true,
     title = {CemoBAM: Advancing Multimodal Emotion Recognition through Heterogeneous Graph Networks and Cross-Modal Attention Mechanisms},
     author = {Nguyen, Nhut Minh and Le, Thu Thuy and Nguyen, Thanh Trung and Phan, Duc Tai and Tran, Anh Khoa and Dang, Duc Ngoc Minh},
     booktitle = {Proceedings of the 2025 Asia-Pacific Network Operations and Management Symposium (APNOMS)},
     doi = {10.23919/apnoms67058.2025.11181320},
     year = {2025},
     month = sep,
     address = {Kaohsiung, Taiwan},
     organization = {National Sun Yat-sen University},
   }

6. APNOMS

   

   ALMUS: Enhancing Active Learning for Object Detection with Metric-Based Uncertainty Sampling

   Duc Tai Phan , Nhut Minh Nguyen , Khang Phuc Nguyen , Tri Pham, and Duc Ngoc Minh Dang

   In Proceedings of the 2025 Asia-Pacific Network Operations and Management Symposium (APNOMS), Sep 2025

   Abs Bib Publication

   Object detection is critical in computer vision but often requires large amounts of labeled data for effective training. Active learning (AL) has emerged as a promising solution to reduce the annotation burden by selecting the most informative samples for labeling. However, existing AL methods for object detection primarily focus on uncertainty sampling, which may not effectively balance the dual challenges of classification and localization. In this study, we explore active learning for object detection, with the objective of optimizing model performance while substantially reducing the demand for annotated data. We propose a novel Active Learning with Metric-based Uncertainty Sampling (ALMUS) that works effectively for the object detection task. This approach prioritizes selecting images containing objects from categories where the model exhibits suboptimal performance, as determined by category-specific evaluation metrics. To balance the annotation budget across different object classes, we propose a dynamic allocation strategy that considers the difficulty of each class and the distribution of object instances within the dataset. This combination of strategies enables our method to effectively address the dual challenges of classification and localization in object detection tasks while still focusing on the rarest and most challenging classes. We conduct extensive experiments on the PASCAL VOC 2007 and 2012 datasets, demonstrating that our method outperforms several active learning baselines. Our results indicate that the proposed approach enhances model performance and accelerates convergence, making it a valuable contribution to the field of active learning in object detection.

   @inproceedings{phan2025ALMUS,
     bibtex_show = true,
     title = {ALMUS: Enhancing Active Learning for Object Detection with Metric-Based Uncertainty Sampling},
     author = {Phan, Duc Tai and Nguyen, Nhut Minh and Nguyen, Khang Phuc and Pham, Tri and Dang, Duc Ngoc Minh},
     booktitle = {Proceedings of the 2025 Asia-Pacific Network Operations and Management Symposium (APNOMS)},
     doi = {10.23919/apnoms67058.2025.11181447},
     year = {2025},
     month = sep,
     address = {Kaohsiung, Taiwan},
     organization = {National Sun Yat-sen University},
   }

7. APNOMS

   

   YOLO-Powered Traffic Sign Detection and OpenStreetMap Integration for Intelligent Navigation

   Duc-Hieu Hoang , Dinh Thuan Nguyen , Quoc Huy Tran , Thanh Trung Nguyen , Nhut Minh Nguyen, and Duc Ngoc Minh Dang

   In Proceedings of the 2025 Asia-Pacific Network Operations and Management Symposium (APNOMS), Sep 2025

   Abs Bib Publication

   This study presents the development of a deep learning-based traffic sign recognition system integrated with OpenStreetMap (OSM) to enhance smart navigation and traffic management. The system utilizes a You Only Look Once (YOLO)based model trained on the Vietnamese Camera-based Traffic Sign Recognition 47 (VCTSR47) dataset to detect and classify Vietnamese traffic signs accurately. A dashboard camera mounted on motorbikes or cars collects real-world traffic data, automatically recognizing traffic signs, extracting Global Positioning System (GPS) coordinates, and storing relevant information in a structured database. The extracted data is integrated into OSM via Application Programming Interfaces (APIs), enabling precise visualization and real-time alerts for critical traffic signs. The system’s performance is rigorously evaluated through various metrics, including mean Average Precision (mAP), Precision, Recall, and Frames Per Second (FPS). Experimental results confirm high recognition accuracy and real-time processing efficiency, demonstrating the feasibility and potential applications of deep learning and digital mapping in modern intelligent transportation systems.

   @inproceedings{hoang2025yolo,
     bibtex_show = true,
     title = {YOLO-Powered Traffic Sign Detection and OpenStreetMap Integration for Intelligent Navigation},
     author = {Hoang, Duc-Hieu and Nguyen, Dinh Thuan and Tran, Quoc Huy and Nguyen, Thanh Trung and Nguyen, Nhut Minh and Dang, Duc Ngoc Minh},
     doi = {10.23919/apnoms67058.2025.11181312},
     booktitle = {Proceedings of the 2025 Asia-Pacific Network Operations and Management Symposium (APNOMS)},
     year = {2025},
     month = sep,
     address = {Kaohsiung, Taiwan},
     organization = {National Sun Yat-sen University},
   }

2024

1. ICTC

   

   Voice-Based Age and Gender Recognition: A Comparative Study of LSTM, RezoNet and Hybrid CNNs-BiLSTM Architecture

   Nhut Minh Nguyen , Thanh Trung Nguyen , Hua Hiep Nguyen, Phuong-Nam Tran, and Duc Ngoc Minh Dang

   In Proceedings of the 2024 15th International Conference on Information and Communication Technology Convergence (ICTC), Sep 2024

   Abs Bib Publication Code

   In this study, we compared three architectures for the task of age and gender recognition from voice data: Long Short-Term Memory networks (LSTM), Hybrid of Convolutional Neural Networks and Bidirectional Long Short-Term Memory (CNNs-BiLSTM), and the recently released RezoNet architecture. The dataset used in this study was sourced from Mozilla Common Voice in Japanese. Features such as pitch, magnitude, Mel-frequency cepstral coefficients (MFCCs), and filter-bank energies were extracted from the voice data for signal processing, and the three architectures were evaluated. Our evaluation revealed that LSTM was slightly less accurate than RezoNet (83.1%), with the hybrid CNNs-BiLSTM (93.1%) and LSTM achieving the highest accuracy for gender recognition (93.5%). However, hybrid CNNs-BiLSTM architecture outperformed the other models in age recognition, achieving an accuracy of 69.75%, compared to 64.25% and 44.88% for LSTM and RezoNet, respectively. Using Japanese language data and the extracted characteristics, the hybrid CNNs-BiLSTM architecture model demonstrated the highest accuracy in both tests, highlighting its efficacy in voice-based age and gender detection. These results suggest promising avenues for future research and practical applications in this field.

   @inproceedings{nguyen2024age-gender,
     bibtex_show = true,
     title = {Voice-Based Age and Gender Recognition: A Comparative Study of LSTM, RezoNet and Hybrid CNNs-BiLSTM Architecture},
     author = {Nguyen, Nhut Minh and Nguyen, Thanh Trung and Nguyen, Hua Hiep and Tran, Phuong-Nam and Dang, Duc Ngoc Minh},
     booktitle = {Proceedings of the 2024 15th International Conference on Information and Communication Technology Convergence (ICTC)},
     pages = {1--1},
     year = {2024},
     publisher = {IEEE},
     doi = {10.1109/ICTC62082.2024.10827387},
   }