Ti Ti Nguyen

Ti Ti Nguyen

Auto-generated profile, based on publicly available information

Ti Ti Nguyen

University of Luxembourg

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

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

Representative Publications

View All Publications
4 October 2019 in IEEE Transactions on Wireless Communications
Verified
Journal Article
Closed access
TNTi Ti Nguyen
Ti Ti Nguyen
VHVu Nguyen Ha
Vu Nguyen Ha
LLLong Bao Le
Long Bao Le
RSRobert Schober
Robert Schober
Data compression (DC) has the potential to significantly improve the computation offloading performance in hierarchical fog-cloud systems. However, it remains unknown how to optimally determine the compression ratio jointly with the computation offloading decisions and the resource allocation. This optimization problem is studied in this paper where we aim to minimize the maximum weighted energy and service delay cost (WEDC) of all users. First, we consider a scenario where DC is performed only at the mobile users. We prove that the optimal offloading decisions have a threshold structure. Moreover, a novel three-step approach employing convexification techniques is developed to optimize the compression ratios and the resource allocation. Then, we address the more general design where DC is performed at both the mobile users and the fog server. We propose three algorithms to overcome the strong coupling between the offloading decisions and the resource allocation. Numerical results show that the proposed optimal algorithm for DC at only the mobile users can reduce the WEDC by up to 65% compared to computation offloading strategies that do not leverage DC or use sub-optimal optimization approaches. The proposed algorithms with additional DC at the fog server lead to a further reduction of the WEDC.

13 January 2019 in IEEE Transactions on Services Computing
Verified
Journal Article
Closed access
NTNguyen Ti Ti
Nguyen Ti Ti
LLLong Bao Le
Long Bao Le
QLQuan Le-Trung
Quan Le-Trung
Intelligent offloading of computation-intensive tasks to a mobile cloud server provides an effective mean to expand the usability of wireless devices and prolong their battery life, especially for low-cost internet-of-things (IoT) devices. However, realization of this technology in multiple-input multiple-output (MIMO) systems requires sophisticated design of joint computation offloading and other network functions such as channel state information (CSI) estimation, beamforming, and resource allocation. In this paper, we study the computation task offloading and resource allocation optimization in MIMO based mobile edge computing systems considering perfect/imperfect-CSI estimation. Our design aims to minimize the maximum weighted energy consumption subject to practical constraints on available computing and radio resources and allowable latency. The optimal and low-complexity algorithms are proposed to solve the underlying mixed integer non-linear problems (MINLP). For the perfect-CSI, we employ bisection search to find the optimal solution. The low-complexity algorithms are developed by decomposing the original optimization problem into the offloading optimization (OP) and power allocation (PA) subproblems and solve them iteratively. Moreover, the difference of convex functions (DC) method is employed to deal with non-convex structure of (PA) subproblems in the imperfect-CSI scenario. Numerical results confirm the advantages of proposed designs over conventional local computation strategies in energy saving and fairness.

16 March 2022 in IEEE Transactions on Mobile Computing
Verified
Journal Article
Closed access
TNTi Ti Nguyen
Ti Ti Nguyen
KNKim-Khoa Nguyen
Kim-Khoa Nguyen
A fine power control policy and beam alignment is required between the base station (BS) and user equipment (UE) to achieve the promising performance of massive multiple input multiple output (MIMO) in millimeter wave (mmWave) communications. However, obtaining the channel state information (CSI) of mmWave - massive MIMO systems is challenging. In this paper, the beam-steering technique is used to estimate the signal strength from the BS to the user. We propose a novel learning framework to determine the suitable beam for a specific user and the transmit power for minimizing the cost including the transmit power and the unsatisfied rate when the channel is unknown. In addition, we address the missing data problem, and then employ the long-short term memory (LSTM) on the temporal processed inputs to select the suitable beam. Furthermore, we design a learning agent to predict the proper transmit power from the transmitted SSBs taking into account the required transmission rate. We then validate the proposed learning framework on the Deep MIMO dataset constructed based on accurate ray-tracing channels. Numerical results show our proposed framework outperforms the state-of-the-art prediction strategies, and approximates the best performance which is obtained when the CSI is available.

8 November 2019 in IEEE Communications Letters
Verified
Journal Article
Closed access
TNTi Ti Nguyen
Ti Ti Nguyen
VHVu Nguyen Ha
Vu Nguyen Ha
LLLong Bao Le
Long Bao Le

8 October 2024 in IEEE Transactions on Communications
Verified
Journal Article
Open access
DTDuc-Dung Tran
Duc-Dung Tran
VHVu Nguyen Ha
Vu Nguyen Ha
SSShree Krishna Sharma
Shree Krishna Sharma
TNTi Ti Nguyen
Ti Ti Nguyen
SCSymeon Chatzinotas
Symeon Chatzinotas
PPPetar Popovski
Petar Popovski
The escalating number of wireless users requiring different services, such as enhanced mobile broadband (eMBB), massive machine-type communications (mMTC), and ultra-reliable low-latency communications (URLLC), has led to exploring non-orthogonal multiplexing methods like heterogeneous non-orthogonal multiple access (H-NOMA). This method allows users demanding divergent services to share the same resources. However, implementing the H-NOMA scheme faces major resource management challenges due to unpredictable interference caused by the random access mechanism of mMTC users. To address this issue, this paper proposes a joint optimization and cooperative multi-agent (MA) deep reinforcement learning-based resource allocation mechanism, aimed at maximizing the energy efficiency (EE) of H-NOMA-based networks. Specifically, this work initially establishes an optimization framework capable of determining the optimal power allocation for any specific sub-channel assignment (SA) setting for all users. Based on that, a cooperative MA double deep Q network (CMADDQN) scheme is carefully designed at the base station to conduct SA among users. In addition, a distributed full learning-based approach using MADDQN for both SA and power allocation is also designed for comparison purposes. Simulation results show that the proposed joint optimization and machine learning method outperforms the solely-learning-based approach and other benchmark schemes in terms of convergence rate and EE performance.

24 June 2025 in IEEE Access
Verified
Journal Article
Open access
TLThanh-Dung Le
Thanh-Dung Le
TNTi Ti Nguyen
Ti Ti Nguyen
VHVu Nguyen Ha
Vu Nguyen Ha
SCSymeon Chatzinotas
Symeon Chatzinotas
PJPhilippe Jouvet
Philippe Jouvet
RNRita Noumeir
Rita Noumeir
Fine-tuning Large Language Models (LLMs) for clinical Natural Language Processing (NLP) poses significant challenges due to domain gap, limited data, and stringent hardware constraints. In this study, we evaluate four adapter techniques—Adapter, Lightweight, TinyAttention, and Gated Residual Network (GRN) - equivalent to Low-Rank Adaptation (LoRA), for clinical note classification under real-world, resource-constrained conditions. All experiments were conducted on a single NVIDIA Quadro P620 GPU (2 GB VRAM, 512 CUDA cores, 1.386 TFLOPS FP32), limiting batch sizes to ≤ 8 sequences and maximum sequence length to 256 tokens. Our clinical corpus comprises only 580 000 tokens, several orders of magnitude smaller than standard LLM pre-training datasets. We fine-tuned three biomedical pre-trained LLMs (CamemBERT-bio, AliBERT, DrBERT) and two lightweight Transformer models trained from scratch. Results show that (i) adapter structures provide no consistent gains when fine-tuning biomedical LLMs under these constraints, and (ii) simpler Transformers, with minimal parameter counts and training times under 6 hours, outperform adapter-augmented LLMs, which required over 1000 GPU-hours. Among adapters, GRN achieved the best metrics (accuracy, precision, recall, F1 = 0.88). These findings demonstrate that, in low-resource clinical settings with limited data and compute, lightweight Transformers trained from scratch offer a more practical and efficient solution than large LLMs, while GRN remains a viable adapter choice when minimal adaptation is needed.