research

2025

1. 

   PBP: Post-training Backdoor Purification for Malware Classifiers

   Dung Thuy Nguyen, Ngoc N Tran , Taylor T Johnson , and 1 more author

   Network and Distributed System Security Symposium (NDSS), 2025

   Abs PDF Code

   This research introduces PBP, a post-training defense for malware classifiers to mitigate backdoor poisoning attacks without relying on specific embedding mechanisms. By regulating batch normalization statistics to correct activation distribution distortions caused by backdoors, PBP effectively restores model performance to a clean state. Experiments show that PBP reduces attack success rates from 100% to nearly 0% using just 1% of training data, significantly outperforming state-of-the-art methods.

2024

1. 

   Backdoor attacks and defenses in federated learning: Survey, challenges and future research directions

   Thuy Dung Nguyen, Tuan Nguyen , Phi Le Nguyen , and 3 more authors

   Engineering Applications of Artificial Intelligence, 2024

   Abs PDF

   Federated learning (FL) has emerged as a promising paradigm for training machine learning models on decentralized data. However, FL is vulnerable to backdoor attacks, where a malicious client injects a backdoor pattern into its local data to manipulate the global model. This paper provides a comprehensive survey of backdoor attacks and defenses in FL. We first introduce the background of FL and backdoor attacks, followed by a detailed review of existing backdoor attacks and defenses in FL. We then discuss the challenges and future research directions in this area. Finally, we provide a summary of the surveyed works and suggest potential research directions for future studies.

2. 

   Iba: Towards irreversible backdoor attacks in federated learning

   Thuy Dung Nguyen, Tuan A Nguyen , Anh Tran , and 2 more authors

   Advances in Neural Information Processing Systems, 2024

   Abs PDF Code

   We propose a novel backdoor attack framework in FL, the Irreversible Backdoor Attack (IBA), that jointly learns the optimal and visually stealthy trigger and then gradually implants the backdoor into a global model. This approach allows the adversary to execute a backdoor attack that can evade both human and machine inspections. Additionally, we enhance the efficiency and durability of the proposed attack by selectively poisoning the model’s parameters that are least likely updated by the main task’s learning process and constraining the poisoned model update to the vicinity of the global model.

3. GAMMA: A universal model for calibrating sensory data of multiple low-cost air monitoring devices

   Anh Duy Nguyen , Thu Hang Phung , Thuy Dung Nguyen, and 3 more authors

   Engineering Applications of Artificial Intelligence, 2024

   Abs PDF

   This study introduces GAMMA, a deep learning-based approach for calibrating data from low-cost air quality monitoring devices, addressing accuracy limitations and the need for device-specific calibration. By leveraging multitask learning to capture shared and device-specific features and employing adversarial training, GAMMA simultaneously calibrates data for multiple devices with improved precision. Implemented in Hanoi, Vietnam, GAMMA demonstrates superior performance, reducing MAE by 60.19% to 74.09% compared to the best baseline.

4. ACML

   Empirical Study of Federated Unlearning: Efficiency and Effectiveness

   Thai-Hung Nguyen , Hong-Phuc Vu , Dung Thuy Nguyen, and 3 more authors

   In Asian Conference on Machine Learning , 2024

   Abs PDF

   Evaluated federated unlearning methods empirically using MNIST, CIFAR-10, and CIFAR-100 and showed that existing methods have difficulty in balancing unlearn time and efficacy.

2023

1. EAAI

   Attentional ensemble model for accurate discharge and water level prediction with training data enhancement

   Anh Duy Nguyen , Viet Hung Vu , Duc Viet Hoang , and 4 more authors

   Engineering Applications of Artificial Intelligence, 2023

   Abs PDF

   This research introduces a novel deep learning-based method for forecasting discharge and water levels, addressing challenges like data scarcity, noise, and underestimation. The approach combines 1D-CNN for feature extraction, LSTM for capturing temporal relationships, ensemble learning for enhanced predictive performance, and Singular-Spectrum Analysis (SSA) to reduce noise. Leveraging an attention mechanism and Linear Exponential (LINEX) loss, the method achieves significant improvements, including up to 40% and 34% gains in the Nash–Sutcliffe Efficiency coefficient for one-step-ahead and multistep-ahead predictions, respectively, compared to existing approaches.

2. 

   FedGrad: Mitigating Backdoor Attacks in Federated Learning Through Local Ultimate Gradients Inspection

   Thuy Dung Nguyen, Anh Duy Nguyen , Thanh-Hung Nguyen , and 4 more authors

   In 2023 International Joint Conference on Neural Networks (IJCNN) , 2023

   Abs PDF Code

   We propose FedGrad, a novel backdoor-resistant defense for FL that is resistant to cutting-edge backdoor attacks, including the edge-case attack, and performs effectively under heterogeneous client data and a large number of compromised clients. FedGrad is designed as a two-layer filtering mechanism that thoroughly analyzes the ultimate layer’s gradient to identify suspicious local updates and remove them from the aggregation process. We evaluate FedGrad under different attack scenarios and show that it significantly outperforms state-of-the-art defense mechanisms. Notably, FedGrad can almost 100% correctly detect the malicious participants, thus providing a significant reduction in the backdoor effect (e.g., backdoor accuracy is less than 8%) while not reducing the main accuracy on the primary task.

2022

1. 

   Feddrl: Deep reinforcement learning-based adaptive aggregation for non-iid data in federated learning

   Nang Hung Nguyen , Phi Le Nguyen , Thuy Dung Nguyen, and 5 more authors

   In Proceedings of the 51st International Conference on Parallel Processing , 2022

   Abs PDF

   To deal with non-IID data, particularly the cluster-skewed data, we propose FedDRL, a novel FL model that employs deep reinforcement learning to adaptively determine each client’s impact factor (which will be used as the weights in the aggregation process). Extensive experiments on a suite of federated datasets confirm that the proposed FedDRL improves favorably against FedAvg and FedProx methods, e.g., up to 4.05% and 2.17% on average for the CIFAR-100 dataset, respectively.

2. ACIIDS

   Image-based contextual pill recognition with medical knowledge graph assistance

   Anh Duy Nguyen , Thuy Dung Nguyen, Huy Hieu Pham , and 2 more authors

   In Asian Conference on Intelligent Information and Database Systems , 2022

   Abs PDF

   The first study to use external prescription data to establish associations between medicines and to classify them using this aiding information. The architecture of PIKA is lightweight and has the flexibility to incorporate into any recognition backbones. The experimental results show that by leveraging the external knowledge graph, PIKA can improve the recognition accuracy from 4.8 % to 31.1% in terms of F1-score, compared to baselines.

2021

1. 

   Joint optimization of charging location and time for network lifetime extension in WRSNs

   Thuy Dung Nguyen, Tuyen Nguyen , Thanh Hung Nguyen , and 2 more authors

   IEEE Transactions on Green Communications and Networking, 2021

   Abs PDF

   Wireless rechargeable sensor networks (WRSNs) have emerged as a potential solution to solve the challenge of prolonging battery-powered sensor networks’ lifetime. In a WRSN, a mobile charger moves around and charges the rechargeable sensors when stopping at charging spots. This paper newly considers a joint optimization of the charging location and the charging time to avoid node failure in WRSNs. We formulate the optimization and propose a solution for that. The proposal includes an algorithm to find a list of potential charging locations and their associated optimal charging time. Moreover, the Q-learning technique is adopted to determine the optimal next charging location among the candidates. We implement the algorithm and conduct experiments to compare it to the related works. The results show that the proposed algorithm outperforms the others in terms of network lifetime. Specifically, the highest performance gap of our proposal to the best algorithm among the others is 8.29 times.