arXiv subjects

Anh T. Nguyen

Publications and source records attributed to Anh T. Nguyen.

6 recordsLinked to original sources

arXiv2026

Lightweight Adaptation of EEG Foundation Models for Stroke Motor Imagery Decoding: Domain Shift and Subject-Level Robustness

Motor imagery (MI) electroencephalography (EEG) decoding could support post-stroke rehabilitation, but models developed on healthy cohorts may not transfer reliably to pathological EEG. We evaluated whether Low-Rank Adaptation (LoRA) can efficiently adapt three pretrained EEG foundation models (i.e., LaBraM-base, REVE-base, and REVE-large) for binary left- versus right-hand MI decoding. Frozen-backbone head-only baselines and LoRA adaptation were evaluated using subject-wise five-fold cross-validation on the PhysioNet EEG Motor Movement/Imagery Dataset and a binary subset of the UET175 dataset comprising 30 stroke participants. On EEGMMIDB, LoRA increased accuracy to 0.822 for LaBraM-base and 0.957 for REVE-base. On UET175, all head-only models performed near chance. With LoRA, LaBraM-base remained near chance (0.499$\pm$0.009), whereas REVE-base reached 0.847$\pm$0.194 and outperformed REVE-large (0.806$\pm$0.178), indicating that increased model capacity alone did not improve stroke-domain adaptation. The strongest stroke configuration, REVE-base LoRA, was further evaluated using within-cohort leave-one-subject-out cross-validation (LOOCV), showing 0.952 mean accuracy, but subject-wise accuracy ranged from 0.586 to 1.000, revealing a small low-performing tail. Zero-shot transfer from EEGMMIDB to UET175 remained near chance (0.464$\pm$0.072). These findings show that healthy-benchmark performance does not ensure transfer to stroke EEG. Translation of EEG foundation models to pseudo-online or real-time rehabilitation BCIs should therefore include target-domain adaptation and subject-level assessment of temporal informativeness, spatial sensitivity, and physiological discriminability.

cs.CE↗

arXiv2026

Robust EEG Functional Connectivity Metrics for Decoding Action Observation Conditions and Observed Actions

Action observation (AO) paradigms probe motor-system engagement, yet the electroencephalographic (EEG) functional connectivity (FC) metrics that best capture AO-related dynamics remain unclear. This pilot study benchmarked five sensor-level FC metrics, including coherence (COH), imaginary coherence (iCOH), phase-locking value (PLV), partial directed coherence (PDC), and spectral Granger causality (SpcG), for decoding AO stimuli in five healthy adults. EEG signals were recorded while participants observed upper-limb actions performed by human or robot agents, as well as non-action control stimuli. Ten motor-area channels were analyzed in the alpha (8-12 Hz) and beta (13-30 Hz) bands. Trial-wise 10 x 10 FC matrices were used as inputs to multiple classifiers for two tasks: (i) six-class AO-condition decoding and (ii) five-class action-type decoding. Across both tasks, metrics robust to volume conduction consistently outperformed their counterparts. iCOH achieved the highest macro-area under the receiver operating characteristic curve (macro-AUC) for most classifiers, with PDC and SpcG showing comparable performance. Graph neural networks (GNNs) provided the most robust and stable results across all FC metrics, while convolutional neural networks and random forests also performed strongly. These findings highlight the importance of suppressing zero-phase-lag interactions and incorporating directed connectivity when characterizing AO-related brain activity. They further demonstrate the ability of GNNs to exploit the inherent graph structure of FC representations, providing practical guidance for selecting connectivity measures and machine learning models in future large-scale studies of action observation.

cs.CE↗

arXiv2026

Expert Merging in Sparse Mixture of Experts with Nash Bargaining

Existing expert merging strategies for Sparse Mixture of Experts (SMoE) typically rely on input-dependent or input-independent averaging of expert parameters, but often lack a principled weighting mechanism. In this work, we reinterpret expert merging through the lens of game theory, revealing cooperative and competitive dynamics among experts. Based on this perspective, we introduce Nash Merging of Experts (NAMEx), a novel framework that incorporates Nash Bargaining into the merging process, enabling more balanced and efficient collaboration among experts. Additionally, we incorporate complex momentum into NAMEx to accelerate expert propagation with theoretical guarantees for convergence. Extensive experiments across language modelling, text classification, image classification, and zero-shot robustness under data corruption show that NAMEx consistently outperforms competing methods while integrating seamlessly with popular MoE architectures. Finally, we demonstrate NAMEx's scalability by applying it to large-scale systems, including Qwen1.5-MoE (14B) and DeepSeek-MoE (16B), where it proves effective in both zero-shot and fine-tuning settings. The code is publicly available at: https://github.com/anh147/NAMEx.

cs.LG↗

arXiv2025

Exploring EEG Responses during Observation of Actions Performed by Human Actor and Humanoid Robot

Action observation (AO) therapy is a promising rehabilitative treatment for motor and language function in individuals recovering from neurological conditions, such as stroke. This pilot study aimed to investigate the potential of humanoid robots to support AO therapy in rehabilitation settings. The brain activity of three healthy right-handed participants was monitored with electroencephalography (EEG) while they observed eight different actions performed by two agents, a human actor and a robot, using their left and right arms. Their event-related spectral perturbations (ERSPs, changes in the spectral power of neural oscillations in response to an event or stimulus, compared to baseline) in sensorimotor regions were analyzed. The single-subject analysis showed variability in ERSP patterns among all participants, including power suppression in sensorimotor mu and beta rhythms. One participant showed stronger responses to "robot" AO conditions than to "human" conditions. Strong and positive correlations in ERSP across all conditions were observed for almost all participants and channels, implying common cognitive processes or neural networks at play in the mirror neuron system during AO. The results support the feasibility of using EEG to explore differences in neural responses to observation of robot- and human-induced actions.

cs.CE↗

arXiv2023

Factors influencing to use of Bluezone

This study aims to understand the main factors and their influence on the behavioral intention of users about using Bluezone. Surveys are sent to users through the Google Form tool. Experimental results through analysis of exploratory factors on 224 survey subjects show that there are 4 main factors affecting user behavior. Structural equation modeling indicates that trust, performance expectations, effort expectations, and social influence have a positive impact on behavioral intention of using Bluezone

cs.HC↗

arXiv2022

VinDr-CXR: An open dataset of chest X-rays with radiologist's annotations

Most of the existing chest X-ray datasets include labels from a list of findings without specifying their locations on the radiographs. This limits the development of machine learning algorithms for the detection and localization of chest abnormalities. In this work, we describe a dataset of more than 100,000 chest X-ray scans that were retrospectively collected from two major hospitals in Vietnam. Out of this raw data, we release 18,000 images that were manually annotated by a total of 17 experienced radiologists with 22 local labels of rectangles surrounding abnormalities and 6 global labels of suspected diseases. The released dataset is divided into a training set of 15,000 and a test set of 3,000. Each scan in the training set was independently labeled by 3 radiologists, while each scan in the test set was labeled by the consensus of 5 radiologists. We designed and built a labeling platform for DICOM images to facilitate these annotation procedures. All images are made publicly available (https://www.physionet.org/content/vindr-cxr/1.0.0/) in DICOM format along with the labels of both the training set and the test set.

eess.IV↗