Tran Hoang Hai

Tran Hoang Hai

Auto-generated profile, based on publicly available information

Tran Hoang Hai

Hanoi University of Science and Technology

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 November 2023 in IEEE Access
Verified
Journal Article
Open access
THTran Hoang Hai
Tran Hoang Hai
VTVu Van Thieu
Vu Van Thieu
TDTran Thai Duong
Tran Thai Duong
HNHong Hoa Nguyen
Hong Hoa Nguyen
Eui-Nam HuhEui-Nam Huh
Eui-Nam Huh
Due to increased reliance on technology and cloud-based services, cyber risks are more common. Advanced persistent threats make it difficult to detect attacks, hence Endpoint Detection and Response (EDR) was adopted in 2013. EDR uses a scanning application on each endpoint machine to monitor and capture events and logs. However, EDR is vulnerable to attacks by malware, so a lightweight malware detector is needed. Image-based malware classification is a technique for classifying malware based on its representative image, but previous studies have not been integrated with EDR. This research aims to integrate EDR with an image-based malware classifier. A basic EDR implementation named Deep Ocean Protection System (DOPS) has been developed with two pre-trained models (Mobilenet V2 and Inception V3) fine-tuned with MalImg and BODMAS datasets. The models were evaluated with the DikeDataset and Mobilenet V2 fine-tuned with BODMAS 4.0.0 performed best in terms of loading and prediction time with a high AUC score of 0.8615. Inception V3 fine-tuned with BODMAS 4.0.0 also achieved a remarkable AUC score of 0.9392. These results show the potential of integrating image-based malware detection with EDR.

23 June 2022 in Sensors
Verified
Journal Article
Open access
PDF file
Tran Trong KhanhTran Trong Khanh
Tran Trong Khanh
THTran Hoang Hai
Tran Hoang Hai
MHMd. Delowar Hossain
Md. Delowar Hossain
Eui-Nam HuhEui-Nam Huh
Eui-Nam Huh
In the era of heterogeneous 5G networks, Internet of Things (IoT) devices have significantly altered our daily life by providing innovative applications and services. However, these devices process large amounts of data traffic and their application requires an extremely fast response time and a massive amount of computational resources, leading to a high failure rate for task offloading and considerable latency due to congestion. To improve the quality of services (QoS) and performance due to the dynamic flow of requests from devices, numerous task offloading strategies in the area of multi-access edge computing (MEC) have been proposed in previous studies. Nevertheless, the neighboring edge servers, where computational resources are in excess, have not been considered, leading to unbalanced loads among edge servers in the same network tier. Therefore, in this paper, we propose a collaboration algorithm between a fuzzy-logic-based mobile edge orchestrator (MEO) and state-action-reward-state-action (SARSA) reinforcement learning, which we call the Fu-SARSA algorithm. We aim to minimize the failure rate and service time of tasks and decide on the optimal resource allocation for offloading, such as a local edge server, cloud server, or the best neighboring edge server in the MEC network. Four typical application types, healthcare, AR, infotainment, and compute-intensive applications, were used for the simulation. The performance results demonstrate that our proposed Fu-SARSA framework outperformed other algorithms in terms of service time and the task failure rate, especially when the system was overloaded.

26 June 2024 in IEEE Access
Verified
Journal Article
Open access
BNBinh Minh Nguyen
Binh Minh Nguyen
TNThieu Nguyen
Thieu Nguyen
QVQuoc-Hien Vu
Quoc-Hien Vu
THTran Huy Hung
Tran Huy Hung
THTran Hoang Hai
Tran Hoang Hai
VTVan-Dang Tran
Van-Dang Tran
This study introduces a brand-new swarm-inspired algorithm dubbed dholes hunting-based optimization (DhoH) based on an animal hunting strategy to solve global optimization problems. The technique is a brilliant idea for simultaneously finding many local minima. The dhole’s hunting strategy is coordinated by members of a swarm, clustering and chasing prey. A clustering approach and finding an optimal global algorithm describe primarily based on gradient approximation. We use four benchmark function datasets to evaluate the DhoH’s performance. We compare the achieved results with several previous research from various well-known algorithms. The comparisons demonstrate that DhoH is better than other meta-heuristic algorithms in most cases and determines high-quality solutions with fewer control parameters. Besides, we also explore the application of DhoH in optimizing the decentralized level of Meta-heuristic Proof of Criteria consensus protocol (MPoC) in Blockchain Network to further demonstrate its potential in multi-dimensional problems. The results show the superior effectiveness of DhoH in terms of computational burden and solution precision compared with the existing optimization techniques in the literature.

9 February 2023 in Research Square
Verified
Preprint
Open access
PDF file
DVDinh-Minh Vu
Dinh-Minh Vu
GNGia Bach Nguyen
Gia Bach Nguyen
HTHoang Hai Tran
Hoang Hai Tran
An alarming number of cyber attacks have been witnessed in recent years, especially during COVID-19 epidemic. Many prevention measures have been proposed as a defense line for a network infrastructure, including Intrusion Detection System (IDS). A variance of IDS using Machine Learning and Deep Learning to detect network anomalies is gaining promising results. However, this approach also poses limitations regarding the indigenous dataset acquisition or the ability to apply a model learned from a benchmark dataset to different network infrastructures. Therefore, this paper proposes a reliable automatic labeling method for a new network dataset, and a Deep Transfer Learning model to detect both known and unknown attacks across different network infrastructures, then compares with other approaches. Network attacks detection using auto-labeling and network-based Deep Learning The obtained results reveal an outstanding performance of Transfer-Learning model, in comparison with non Transfer-Learning method, on two benchmark datasets (NSL-KDD, CIC-IDS2017) and a self-captured simulated network dataset using the auto-labeling approach. Furthermore, the model is verified to retain both new and old knowledge after the transfer learning process, which has not been mentioned in other studies, focusing only on learning new knowledge ability.

1 November 2021
Verified
Conference Paper
Closed access
NANguyen The Anh
Nguyen The Anh
LHLe Huy Hoang
Le Huy Hoang
Vu Dinh MinhVu Dinh Minh
Vu Dinh Minh
THTran Hoang Hai
Tran Hoang Hai
Applying machine learning techniques to Intrusion Detection System (IDS) is one efficient method to detect existing and new types of network attacks with minimal human supervision and great accuracy. The classification performance of such systems depend heavily on the dataset they are trained on. Most currently deployed datasets suffer from a high imbalance between normal network traffic and anomaly network traffic, which limits the system’s ability to detect minority classes with high accuracy. In this paper, we propose a new model for generating and collecting anomaly network traffic, which is used in conjunction with CICIDS2017 [1] dataset to create a new dataset namely BKIDset, which mitigates the class imbalance problem encountered in CICIDS2017 dataset and introduces additional types of denial-of-service attacks. The paper also proposes different machine learning models trained on BKIDset, that achieve a high accuracy score of 99%.