Specific Tasks Research Articles

The ‘Sandwich’ meta-framework for architecture agnostic deep privacy-preserving transfer learning for non-invasive brainwave decoding

Objective. Machine learning has enhanced the performance of decoding signals indicating human behaviour. Electroencephalography (EEG) brainwave decoding, as an exemplar indicating neural activity and human thoughts non-invasively, has been helpful in neural activity analysis and aiding paralysed patients via brain-computer interfaces. However, training machine learning algorithms on EEG encounters two primary challenges: variability across data sets and privacy concerns using data from individuals and data centres. Our objective is to address these challenges by integrating transfer learning for data variability and federated learning for data privacy into a unified approach.Approach. We introduce the 'Sandwich' as a novel deep privacy-preserving meta-framework combining transfer learning and federated learning. The 'Sandwich' framework comprises three components: federated networks (first layers) that handle data set differences at the input level, a shared network (middle layer) learning common rules and applying transfer learning techniques, and individual classifiers (final layers) for specific brain tasks of each data set. This structure enables the central network (central server) to benefit from multiple data sets, while local branches (local servers) maintain data and label privacy.Main results. We evaluated the 'Sandwich' meta-architecture in various configurations using the BEETL motor imagery challenge, a benchmark for heterogeneous EEG data sets. Compared with baseline models likeShallow ConvNetandEEGInception, our 'Sandwich' implementations showed superior performance. The best-performing model, the Inception SanDwich with deep set alignment (Inception-SD-Deepset), exceeded baseline methods by 9%.Significance. The 'Sandwich' framework demonstrates advancements in federated deep transfer learning for diverse tasks and data sets. It outperforms conventional deep learning methods, showcasing the potential for effective use of larger, heterogeneous data sets with enhanced privacy. In addition, through its diverse implementations with various backbone architectures and transfer learning approaches, the 'Sandwich' framework shows the potential as a model-agnostic meta-framework for decoding time series data like EEG, suggesting a direction towards large-scale brainwave decoding by combining deep transfer learning with privacy-preserving federated learning.

Transforming Physiology and Healthcare through Foundation Models.

Recent developments in artificial intelligence (AI) may significantly alter physiological research and healthcare delivery. Whereas AI applications in medicine have historically been trained for specific tasks, recent technological advances have produced models trained on more diverse datasets with much higher parameter counts. These new, "foundation" models raise the possibility that more flexible AI tools can be applied to a wider set of healthcare tasks than in the past. This review describes how these newer models differ from conventional task-specific AI, which relies heavily on focused datasets and narrow, specific applications. By examining the integration of AI into diagnostic tools, personalized treatment strategies, biomedical research, and healthcare administration, we highlight how these newer models are revolutionizing predictive healthcare analytics and operational workflows. In addition, we address ethical and practical considerations associated with the use of foundation models by highlighting emerging trends, calling for changes to existing guidelines, and emphasizing the importance of aligning AI with clinical goals to ensure its responsible and effective use.

Cheat, cheat, repeat: On the consistency of dishonest behavior in structurally comparable situations.

A fundamental assumption about human behavior forming the backbone of trait theories is that, to some extent, individuals behave consistently across structurally comparable situations. However, especially for unethical behavior, the consistency assumption has been severely questioned, at least from the early 19th century onward. We provide a strict test of the consistency assumption for a prominent instance of unethical behavior-dishonesty-in a large (N = 1,916) and demographically diverse sample. Dishonest behavior was measured three times-up to 3 years apart-using different variants of well-established, incentivized cheating paradigms. A key advantage of these paradigms is that lying is individually profitable but not self-incriminating. Besides varying the specific task at hand, we experimentally varied the nature of incentives (i.e., money vs. avoiding tedious work) as well as their magnitude across measurement occasions. The consistency of dishonest behavior was estimated using a newly developed statistical model. Results showed strong consistency of dishonest behavior across contexts in most cases. Furthermore, theoretically relevant personality traits (i.e., Honesty-Humility and the Dark Factor) yielded meaningful relations both with dishonesty and indeed its consistency. Thus, contrary to long-standing assumptions, there is notable consistency in dishonest behavior that can be attributed to underlying dispositional factors. Overall, the current findings have important implications for the theoretical understanding of dishonest behavior by providing strong evidence for (dis)honesty as a trait as well as for practice (e.g., honesty interventions). Moreover, the newly developed statistical approach can serve future research across scientific fields. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

COCONUT PREDICTIVE ANALYSIS USING MACHINE LEARNING

“Coconut Predictive Analysis” focuses on leveraging advanced machine learning techniques to predict the suitability of coconuts for oil production. The project employs transfer learning with the Inception V3 model, a state-of-the-art convolutional neural network (CNN), to analyze complex patterns in coconut-related data, such as images of coconuts. Traditional methods for determining coconut suitability for oil production rely on manual inspection, which can be time-intensive, subjective, and inconsistent. By utilizing transfer learning, this project capitalizes on pre-trained Inception V3 models, which extract meaningful features from large-scale datasets like ImageNet, and fine-tunes them for the specific task of classifying dry coconuts based on their suitability for oil extraction. The fine-tuning process involves freezing initial layers to retain general features while optimizing later layers to improve prediction accuracy for coconut-specific characteristics such as texture, colour, and surface patterns. To ensure accessibility and usability, a Flask web application is developed. This application enables users to upload images of coconuts and receive real-time predictions on their suitability for oil production. The backend integrates the trained model to process inputs, extract features, and generate predictions, making it a scalable and practical solution for agricultural and industrial applications. This project demonstrates the potential of transfer learning in enhancing agricultural and processing efficiency by reducing training time and computational costs while providing accurate and actionable predictions. By combining cutting-edge deep learning with a user-friendly web interface, it bridges the gap between advanced AI models and practical applications, supporting decision-making for farmers, coconut processing units, and industry stakeholders. Key Words: Inception V3, fine-tuning, CNN, Flask, coconut analysis, oil production suitability

PE-GCL: Advancing pesticide ecotoxicity prediction with graph contrastive learning.

Ecotoxicity assessments, which rely on animal testing, face serious challenges, including high costs and ethical concerns. Computational toxicology presents a promising alternative; nevertheless, existing predictive models encounter difficulties such as limited datasets and pronounced overfitting. To address these issues, we propose a framework for predicting pesticide ecotoxicity using graph contrastive learning (PE-GCL). By pre-training on large-scale unlabeled compounds, the PE-GCL captured the intrinsic regulation of molecules. This knowledge is then transferred to specific downstream tasks, thereby enhancing the model generalization in scenarios with small sample sizes. Performance evaluation showed that the PE-GCL outperformed traditional supervised models across most prediction tasks, whereas independent external validation confirmed its superior predictive accuracy for unseen data. Furthermore, interpretability was incorporated to elucidate potential correlations between ecotoxicity and molecular substructures. The trained models were deployed on a publicly accessible web server (https://dpai.ccnu.edu.cn/PERA/) to facilitate the use of the proposed framework.

BoundMPC: Cartesian path following with error bounds based on model predictive control in the joint space

This work introduces the BoundMPC strategy, an innovative online model-predictive path-following approach for robot manipulators. This joint-space trajectory planner allows the following of Cartesian reference paths in the end-effector’s position and orientation, including via-points, within the desired asymmetric bounds of the orthogonal path error. These bounds encode the obstacle-free space and additional task-specific constraints in Cartesian space. Contrary to traditional path-following concepts, BoundMPC purposefully deviates from the Cartesian reference path in position and orientation to account for the robot’s kinematics, leading to more successful task executions for Cartesian reference paths. Furthermore the simple reference path formulation is computationally efficient and allows for replanning during the robot’s motion. This feature makes it possible to use this planner for dynamically changing environments and varying goals. The flexibility and performance of BoundMPC are experimentally demonstrated by five scenarios on a 7-DoF Kuka LBR iiwa 14 R820 robot. The first scenario shows the transfer of a larger object from a start to a goal pose through a confined space where the object must be tilted. The second scenario deals with grasping an object from a table where the grasping point changes during the robot’s motion, and collisions with other obstacles in the scene must be avoided. The adaptability of BoundMPC is showcased in scenarios such as the opening of a drawer, the transfer of an open container, and the wiping of a table, where it effectively handles task-specific constraints. The last scenario highlights the possibility of accounting for collisions with the entire robot’s kinematic chain. The code is readily available at https://github.com/TU-Wien-ACIN-CDS/BoundMPC , inspiring you to explore its potential and adapt it to your specific robotic tasks.

Machine Learning Methods for Weather Forecasting: A Survey

Weather forecasting, a vital task for agriculture, transportation, energy, etc., has evolved significantly over the years. Comprehensive surveys play a crucial role in synthesizing knowledge, identifying trends, and addressing emerging challenges in this dynamic field. In this survey, we critically examines machine learning (ML)-based weather forecasting methods, which demonstrate exceptional capability in handling complex, high-dimensional datasets and leveraging large volumes of historical and real-time data, enabling the identification of subtle patterns and relationships among weather variables. Research on specific tasks such as global weather forecasting, downscaling, extreme weather prediction, and how to combine machine learning methods with physical principles are very active in the current field. However, several unresolved or challenging issues remain, including the interpretability of models and the ability to predict rare weather events. By identifying these gaps, this research provides a roadmap for advancing machine learning-based weather forecasting techniques to complement and enhance weather prediction results.

Diagnosis of Parkinson’s disease by eliciting trait-specific eye movements in multi-visual tasks

BackgroundParkinson’s Disease (PD) is a neurodegenerative disorder, and eye movement abnormalities are a significant symptom of its diagnosis. In this paper, we developed a multi-task driven by eye movement in a virtual reality (VR) environment to elicit PD-specific eye movement abnormalities. The abnormal features were subsequently modeled by using the proposed deep learning algorithm to achieve an auxiliary diagnosis of PD.MethodsWe recruited 114 PD patients and 125 healthy controls and collected their eye-tracking data in a VR environment. Participants completed a series of specific VR tasks, including gaze stability, pro-saccades, anti-saccades, and smooth pursuit. After the tasks, eye movement features were extracted from the behaviors of fixations, saccades, and smooth pursuit to establish a PD diagnostic model.ResultsThe performance of the models was evaluated through cross-validation, revealing a recall of 97.65%, an accuracy of 92.73%, and a receiver operator characteristic area under the curve (ROC-AUC) of 97.08% for the proposed model.ConclusionWe extracted PD-specific eye movement features from the behaviors of fixations, saccades, and smooth pursuit in a VR environment to create a model with high accuracy and recall for PD diagnosis. Our method provides physicians with a new auxiliary tool to improve the prognosis and quality of life of PD patients.

Simulation of semiconductor wafer dicing induced faults on chips and their application as augmentation method for a deep learning based visual inspection system

Abstract In semiconductor wafer dicing, one particular area of interest is the process of visual inspection to detect manufacturing defects that occur throughout the manufacturing process. Emerging defect patterns are typically in the micrometer range, translating to barely visible defects in pixel size on high-resolution imagery. The availability of rarely occurring defects is generally limited due to the labor-intensive nature of the related, highly specialized annotation task. Therefore, this contribution proposes a hybrid system for wafer and chip image data synthesis utilizing wafer and dicing path templates for the synthetic generation of large quantities of labeled flawless and, rarely, faulty chip and dicing street imagery. These are utilized for subsequent deep learning based defect detection and classification by employing a residual neural network as the core classifier for our visual inspection system. Our results show promising prospects when the original image data are supplemented with synthesized images by creating so-called composite data sets. Compared to the system’s baseline on the original data set, an F1-score-based relative improvement of up to 3.98 times was achieved. Furthermore, a novel synthetic-composite leave-one-out cross-validation (SC-LOOCV) method is proposed as a means to analyze the quality of our synthesized data for each specific wafer type. Based on these experiments, we scored a relative improvement of up to 5.99. For all our wafer types, overall relative improvement factors of 1.99 (composite) and 2.83 (SC-LOOCV) highlight the benefits of our realized system.

Major individual and regional variations in unit entrainment by oscillations of different frequencies

In vitro studies have shown that a neuron’s electroresponsive properties can predispose it to oscillate at specific frequencies. In contrast, network activity in vivo can entrain neurons to rhythms that their biophysical properties do not predispose them to favor. However, there is limited information on the comparative frequency profile of unit entrainment across brain regions. Therefore, this study aimed to characterize the frequency profile of unit entrainment in cortex, thalamus, striatum, and basolateral amygdala (BLA) in rats of either sex. Neurons recorded simultaneously in a given brain region and behavioral state generally had very similar frequency profiles of unit entrainment. While cortical, striatal, and thalamic neurons were more strongly entrained by low than high local field potential (LFP) frequencies, increases in the power of these oscillations were linked to decreased firing rates for low frequencies versus increased firing rates for high frequencies. Deviating from this general trend, BLA neurons were more strongly entrained by high gamma than all other frequency bands in all subjects and states. By contrast, neurons in other regions displayed marked inter-individual variability. That is, although neurons in some regions had exceptionally high entrainment values in particular frequency bands, these were not observed consistently across rats. Based on these findings, some might infer that oscillations play a minor role or that different oscillatory patterns can support the same functions. Alternatively, the oscillations critical to brain function could be those not investigated here, namely those arising transiently in response to specific task variables or contexts. Perhaps those are less susceptible to genetic variations. While our findings do not allow us to determine which explanation is correct, they do highlight the perils of averaging.

Trustworthy and efficient project scheduling in IIoT based on smart contracts and edge computing

To facilitate flexible manufacturing, modern industries have incorporated numerous modular operations such as multi-robot services which can be expediently arranged or offloaded to other production resources. However, complex manufacturing projects often consist of multiple tasks with fixed sequences, posing a significant challenge for smart factories in efficiently scheduling limited robot resources to complete specific tasks. Additionally, when projects span across factories, ensuring faithful execution of contracts becomes another challenge. In this paper, we propose a modified combinatorial auction method combined with blockchain and edge computing technologies to organize project scheduling. Firstly, we transform efficient resource scheduling into a resource-constrained multi-project scheduling problem (RCPSP). Subsequently, the solution integrates combinatorial auction with random sampling (CA-RS) into smart contracts. Alongside security analysis, simulations are conducted using real data sets. The results indicate that the suggested CA-RS approach significantly enhances efficiency and security in resource arrangement within the industrial Internet of Things compared to baseline algorithms.

Differences in User Perception of Artificial Intelligence-Driven Chatbots and Traditional Tools in Qualitative Data Analysis

Qualitative data analysis (QDA) tools are essential for extracting insights from complex datasets. This study investigates researchers’ perceptions of the usability, user experience (UX), mental workload, trust, task complexity, and emotional impact of three tools: Taguette 1.4.1 (a traditional QDA tool), ChatGPT (GPT-4, December 2023 version), and Gemini (formerly Google Bard, December 2023 version). Participants (N = 85), Master’s students from the Faculty of Electrical Engineering and Computer Science with prior experience in UX evaluations and familiarity with AI-based chatbots, performed sentiment analysis and data annotation tasks using these tools, enabling a comparative evaluation. The results show that AI tools were associated with lower cognitive effort and more positive emotional responses compared to Taguette, which caused higher frustration and workload, especially during cognitively demanding tasks. Among the tools, ChatGPT achieved the highest usability score (SUS = 79.03) and was rated positively for emotional engagement. Trust levels varied, with Taguette preferred for task accuracy and ChatGPT rated highest in user confidence. Despite these differences, all tools performed consistently in identifying qualitative patterns. These findings suggest that AI-driven tools can enhance researchers’ experiences in QDA while emphasizing the need to align tool selection with specific tasks and user preferences.

Enhancing Intrusion Detection Systems with Adaptive Neuro-Fuzzy Inference Systems

Network security has become increasingly critical in recent years. Among the various aspects of network security and considering several approaches to network security, intrusion detection systems (IDSs) have gained considerable attention. The prominence of this factor, among other factors of network security, is due to its ability to address the complex and uncertain nature of security breaches. Whenever data flow over the network, precise categorization of normal and malicious data is necessary. Past IDS systems lack precise categorization. Thus, the present study focuses on the use of the adaptive neuro-fuzzy inference system (ANFIS) as a classifier to categorize network instances into malicious types and normal behavior. Using the KDD99 dataset, the performance of ANFIS is evaluated and compared with that of traditional machine learning models such as decision trees and multilayer perceptrons. Through experimentation with different membership functions, such as Gaussian, triangular, bell-shaped, and sigmoidal functions, Gaussian functions are identified as optimal for this specific task. The results underscore the effectiveness of ANFIS, leveraging the strengths of both artificial neural networks (ANNs) and fuzzy reasoning systems. ANFIS demonstrates superior capabilities in understanding nonlinear interaction patterns, adapting to evolving threats, and facilitating rapid learning in intrusion detection applications.

Maternal touch in object- and nonobject-oriented play interactions: A longitudinal study at 7 and 12 months.

Social touch is a crucial part of how mothers interact with their infants, with different touch types serving distinct purposes in these exchanges. However, there is still a limited understanding of how mothers' touch behavior adapts to specific interactive tasks, particularly throughout infancy. To address this gap, we observed mother-infant dyads at 7 and 12 months during three structured social play tasks: (a) play with objects, (b) play without objects, and (c) play with a difficult object. Using an adapted version of the Ordinalized Maternal Touch Scale, we categorized every touch performed by the mother. The effect of the infant's age and play tasks on the proportion of time mothers touch their infants was evaluated using Bayesian beta mixed models, taking into account both the total quantity and the Ordinalized Maternal Touch Scale touch categories. Results showed that (a) the frequency of maternal touch is prevalent in dyadic interactions and lowered in triadic object play; (b) mothers used affectionate, static, and playful touch categories more often in dyadic play tasks; (c) in triadic play task, mothers used object-mediated touch more frequently; (d) the total frequency of maternal touch decreased across infant age, which was primarily due to a decrease in static and object-mediated touch; and (e) maternal touch varies depending on the complexity of object play task. Our findings suggested that the developmental trajectory of maternal touch behavior is modulated by the infant's evolving needs and the different challenges in object versus nonobject play tasks. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

Efficient Multi-Task Training with Adaptive Feature Alignment for Universal Image Segmentation.

Universal image segmentation aims to handle all segmentation tasks within a single model architecture and ideally requires only one training phase. To achieve task-conditioned joint training, a task token needs to be used in the multi-task training to condition the model for specific tasks. Existing approaches generate the task token from a text input (e.g., "the task is panoptic"). However, such text-based inputs merely serve as labels and fail to capture the inherent differences between tasks, potentially misleading the model. In addition, the discrepancy between visual and textual modalities limits the performance gains in existing text-involved segmentation models. Nevertheless, prevailing modality-alignment methods rely on large-scale uni-modal encoders for both modalities and an extremely large amount of paired data for training, and therefore it is hard to apply these existing models to lightweight segmentation models and resource-constrained devices. In this paper, we propose Adaptive Feature Alignment (AFA) integrated with a learnable task token to address these issues. The learnable task token automatically captures inter-task differences from both image features and text queries during training, providing a more effective and efficient solution than a predefined text-based token. To efficiently align the two modalities without introducing extra complexity, we reconsider the differences between a text token and an image token and replace image features with class-specific means in our proposed AFA. We evaluate our model performance on the ADE20K and Cityscapes datasets. Experimental results demonstrate that our model surpasses baseline models in both efficiency and effectiveness, achieving state-of-the-art performance among segmentation models with a comparable amount of parameters.

Analysis of reticulated dome with universal connector

In order to investigate the stress-strain state of a 4 m diameter reticulated dome model, four series of glued wooden rods were prepared for centric and eccentric compression tests. The tests were carried out in the laboratory of the Department of Metal Wood and Plastic Structures. The stresses at the specific points of the elements were determined through the deformations using the strain gauges. The feature of the prismatic specimens was the stress concentrators in the bearing areas in the form of holes for the arrangement of universal connectors. These areas were also reinforced with steel sleeves. The general conclusion of the study is the high load bearing capacity of the tested samples. The destruction of the samples occurred in the bearing zone due to wood crushing. The next research tasks will be to optimise the dimensions of the elements and test the dome model. The cross-section of the elements, in addition to providing the load-bearing capacity, is also influenced by the need to obtain certain thermal characteristics of the enclosure, i.e. the dome elements should have dimensions that allow placing a layer of an effective insulation in their plane. A special task is to choose a roofing that can be considered only as a part of the constant load on the load-bearing system or as a continuous shell that unbinds the frame.

Applying MLP-Mixer and gMLP to Human Activity Recognition.

The development of deep learning has led to the proposal of various models for human activity recognition (HAR). Convolutional neural networks (CNNs), initially proposed for computer vision tasks, are examples of models applied to sensor data. Recently, high-performing models based on Transformers and multi-layer perceptrons (MLPs) have also been proposed. When applying these methods to sensor data, we often initialize hyperparameters with values optimized for image processing tasks as a starting point. We suggest that comparable accuracy could be achieved with fewer parameters for sensor data, which typically have lower dimensionality than image data. Reducing the number of parameters would decrease memory requirements and computational complexity by reducing the model size. We evaluated the performance of two MLP-based models, MLP-Mixer and gMLP, by reducing the values of hyperparameters in their MLP layers from those proposed in the respective original papers. The results of this study suggest that the performance of MLP-based models is positively correlated with the number of parameters. Furthermore, these MLP-based models demonstrate improved computational efficiency for specific HAR tasks compared to representative CNNs.

Exploring Brain Size Asymmetry and Its Relationship with Predation Risk Among Chinese Anurans.

Brain size asymmetry differs considerably across species, including humans, vertebrates, and invertebrates. The subtle structural, functional, or size differences between the two brain sides are associated with processing specific cognitive tasks. To evaluate the differences between the sizes of the left and right sides of the whole brain and brain regions and the effect of predation risk (i.e., snake density) on brain size asymmetry among Chinese anurans, we compared the differences between the left and right hemisphere sizes of the whole brain and brain regions among anuran species and analyzed the correlations between the predation risk and size asymmetry index of the brain and brain regions. We found that when one side of the brain was consistently larger than the other, there was a significant difference between the sizes of the left and right sides of the brain and brain regions, displaying directional asymmetry of the whole brain and brain regions. We also found that total brain size was positively correlated with the size asymmetry index of the olfactory bulb and optic tecta when the left hemispheres of the whole brain and brain regions were larger than the right ones. Meanwhile, the index of telencephalon size asymmetry was positively correlated with predation risk when the right hemispheres of the brain and brain regions were larger than the left ones. However, there were non-significant differences between the sizes of the left and right sides of the brain and brain regions across 99 species of anurans. Our findings suggest that an increased predation risk linked to sociality is likely to drive an increase in right telencephalon size.

Good Practices for Health Technology Assessment Guideline Development: A Report of the Health Technology Assessment International, HTAsiaLink, and ISPOR Special Task Force.

Health technology assessment (HTA) guidelines are intended to support the successful implementation of HTA by enhancing consistency and transparency in concepts, methods, processes, and use, thereby enhancing the legitimacy of the decision-making process. This report lays out good practices and practical recommendations for developing or updating HTA guidelines to ensure successful implementation. The task force was established in 2022 and comprised experts and academics from various geographical regions, each with substantial experience in developing HTA guidelines for national health policy making. Literature reviews and key informant interviews were conducted to inform these good practices. Stakeholder consultations, open peer reviews, and expert opinions validated the recommendations. A series of teleconferences among task force members was held to iteratively refine the report. The recommendations cover six key aspects throughout the guideline development cycle: (1) setting objectives, scope, and principles of the guideline, (2) building a team for a quality guideline, (3) defining a stakeholder engagement plan, (4) developing content and utilizing available resources, (5) putting in place appropriate institutional arrangements, and (6) monitoring and evaluating guideline success. This report presents a set of resources and context-appropriate practices for developing or updating HTA guidelines. Across all contexts, the recommendations emphasize transparency, building trust among stakeholders, and fostering a culture of ongoing learning and improvement. The report recommends timing development and revision of guidelines according to the HTA landscape and pace of HTA institutionalization. Because HTA is increasingly used to inform different kinds of decision making in a variety of country contexts, it will be important to continue to monitor lessons learned to ensure the recommendations remain relevant and effective.

Narrative mobile video game-based cognitive training to enhance frontal function in patients with mild cognitive impairment

Although cognitive training has been proposed as a possible therapeutic modality for mild cognitive impairment (MCI), most serious games focus on specific tasks. This study aimed to investigate the feasibility and efficacy of narrative video game-based cognitive intervention for MCI. A four-week (± 1-week) mobile game intervention was given to 17 MCI participants (mean age (SD) = 72.8(4.75)). At baseline and post-intervention, the participants received neuropsychological tests and a depression scale. Frontal function was assessed using the Corsi block-tapping test, Color Word Stroop Test, Controlled Oral Word Association Test, Digit Symbol Coding, and Trail Making Test-Elderly’s Version; depression was assessed using the Geriatric Depression Scale. User’s compliance and gaming experience were also evaluated. MCI patients showed significant improvements in frontal function, particularly in Digit Symbol Coding (mean ± SD, 0.47 ± 0.49, p = 0.007) and phonemic fluency (mean ± SD, 0.39 ± 0.55, p = 0.024). Each frontal subtest’s mean z-score was increased (mean ± SD, 0.44 ± 0.38, p = 0.008). Block span and depression scale remained unchanged. High adherence rates (122.35%) and favorable feedback on the gaming experience indicated that the game intervention’s usability boosted patients’ motivation and engagement. Our findings demonstrate that narrative game-based cognitive intervention was not only beneficial but also enjoyable for elderly MCI.