Earticle

적응적 학습 시스템은 학습자의 학습 요구에 따라서 학습 자료를 적응적으로 제공해주는 시스템을 의미한다. 적응적 학습 시스템은 전문가 모델, 수업 모델, 학습자 모델로 구성되어있다. 전문가 모델은 가르치는 정보를 저장하고 있다. 학습자 모델은 학생들의 학습 정보와 학습 이력에 대한 데이터를 저장한다. 수업 모델은 실제 학습자에게 필요한 학습 자료를 제공해주는 모델이다. 본 논문에서는 학습자 프로파일 정보를 통하여 학습자 모델을 구성하였으며, 동적 시나리오 구축을 통하여 수업 모델을 구성하였다. 이후 학습자의 프로파일 정보 기반의 동적 시나리오를 구축해줌으로써 학습자에게 적응적으로 학습 콘텐츠를 제공해주는 시스템을 개발하였다. 마지막으로 시스템에 대한 만족도 결과는 88%로 높은 만족도를 보였다.

Adaptive learning system means a system that provides adaptively learning materials according to the learning needs of learners. It consists of expert model, instructional model and student model. Expert model is that stores information which is to be taught. Student model stores the data of learning history and learning information of students. Instructional model provides necessary learning materials for actual leaners. This paper has constructed student model through learner’s profile information and instructional model through dynamic scenario construction. After that, We have developed adaptively to provide learning to learners by constructing suitable dynamic scenario based on learners profile information. In the end, satisfaction result about this system showed a high degree of satisfaction and 88%.

최근 인공지능을 구현하기 위한 기술들이 보편화되면서 특히, 기계 학습이 폭넓게 사용되고 있다. 기계 학습은 대량 의 데이터를 수집하고 일괄적으로 처리하며 최종 조치를 취할 수 있는 통찰력을 제공하나, 작업의 효과가 즉시 학습 과정에 통합되지는 않는다. 본 연구에서는 비즈니스의 큰 이슈로서 실시간 데이터 분석의 성능을 개선하기 위한 적응형 학습 모델 을 제안하였다. 적응형 학습은 데이터세트의 복잡성에 적응하여 앙상블을 생성하고 알고리즘은 샘플링 할 최적의 데이터 포인트를 결정하는데 필요한 데이터를 사용한다. 6개의 표준 데이터세트를 대상으로 한 실험에서 적응형 학습 모델은 학습 시간과 정확도에서 분류를 위한 단순 기계 학습 모델보다 성능이 우수하였다. 특히 서포트 벡터 머신은 모든 앙상블의 후단 에서 우수한 성능을 보였다. 적응형 학습 모델은 시간이 지남에 따라 다양한 매개변수들의 변화에 대한 추론을 적응적으로 업데이트가 필요한 문제에 폭넓게 적용될 수 있을 것으로 기대한다.

Recently, as technologies for realizing artificial intelligence have become more common, machine learning is widely used. Machine learning provides insight into collecting large amounts of data, batch processing, and taking final action, but the effects of the work are not immediately integrated into the learning process. In this paper proposed an adaptive learning model to improve the performance of real-time stream analysis as a big business issue. Adaptive learning generates the ensemble by adapting to the complexity of the data set, and the algorithm uses the data needed to determine the optimal data point to sample. In an experiment for six standard data sets, the adaptive learning model outperformed the simple machine learning model for classification at the learning time and accuracy. In particular, the support vector machine showed excellent performance at the end of all ensembles. Adaptive learning is expected to be applicable to a wide range of problems that need to be adaptively updated in the inference of changes in various parameters over time.

본 연구는 대구시에 위치한 중학교 1개교에 재학 중인 266명의 중학생을 대상으로 인공지능 기반 적응형 학습 프로그램을 제공하는 온라인학습기기가 학생들의 학업성취도에 미친 영향을 무작위통제실험 방식으로 분석하였다. 기초 및 기말 진단고사 성적 변화를 분석한 결과, 통계적으로 유의미한 온라인학습기기의 평균처치효과는 관찰되지 않았다. 하지만 분위회귀분석을 통해 하위 10% 학생들의 학업성과가 0.358 표준편차만큼 통계적으로 유의미하게 증가함을 확인하였다. 처치집단 학생들의 저조한 프로그램 출석률과 수행률을 고려할 때, 인공지능 기반 적응형 학습 프로그램의 수용성을 높이기 위한 다양한 방법을 탐색하는 연구가 선행될 필요가 있다.

We investigate the impact of artificial intelligence-based adaptive learning program provided to 266 students in one middle school in Daegu city, South Korea. While examining the difference between the baseline and endline academic performance, we do not find the average treatment effect of the adaptive learning program. However, the quantile regression analysis by deciles shows that there is 0.358 standard deviation improvement for the first decile group. Given the very low attendance and task completion rates of the adaptive learning program, how to increase the adoption for this type of education technology is a prerequisite for the future research

본 연구의 목적은 2015년부터 2025년까지 국내 대학교육을 대상으로 수행된 인공지능 기반 적응형 학습 관 련 학술논문을 주제별로 고찰하여 연구 동향과 교육적 효과를 분석하는 것이다. Arksey와 O’Malley가 제시하고 Levac 등이 보완한 5단계 주제범위 문헌고찰 방법론을 적용하였으며, PRISMA-ScR 지침에 따라 연구를 수행하였 다. 국내 데이터베이스인 RISS, DBpia, KISS, KCI에서 ‘인공지능’, ‘적응형 학습’, ‘맞춤형 학습’, ‘대학’, ‘고등교육’ 등의 주요어를 조합하여 2015–2025년 사이 게재된 논문을 검색한 결과, 총 220편 중 선정 기준을 충족한 12편이 최종 분석 대상이 되었다. 선정된 연구의 주요 학습성과로는 학업성취, 학습동기, 학습몰입, 자기주도학습능력, 디 지털 리터러시, 학업 자기효능감 등이 보고되었고, 대부분의 연구에서 AI 기반 적응형 학습이 학업성취를 유의하 게 향상시키는 것으로 나타났다. 결론적으로 AI 기반 적응형 학습은 국내 대학교육에서 학업성취와 정의적·메타인 지적 학습성과를 동시에 향상시킬 수 있는 유망한 교수학습 전략이며, 저성취 학습자 지원에 효과적이다.

This study analyzed research trends and educational effects by conducting a scoping review of academic papers on artificial intelligence (AI)-based adaptive learning implemented in Korean higher education between 2015 and 2025. The five-stage thematic scoping review methodology proposed by Arksey and O’Malley and refined by Levac et al. was applied, in accordance with the PRISMA-ScR guidelines. Academic articles published between 2015 to 2025 were identified in searches of four Korean databases (RISS, DBpia, KISS, and KCI) using combinations of the keywords ‘artificial intelligence’, ‘adaptive learning’, ‘personalized learning’, ’university’, and ‘higher education’. Of the 220 records initially retrieved, 12 studies met the predefined inclusion criteria and were included in the final analysis. The principal learning outcomes most frequently reported in these studies were academic achievement, learning motivation, learning flow, self-directed learning ability, digital literacy, and academic self-efficacy. Across most studies, AI-based adaptive learning was associated with statistically significant improvements in academic performance. Overall, the findings indicate that AI-based adaptive learning is a promising pedagogical strategy in Korean university education, with the potential to enhance both academic achievement and metacognitive learning outcomes, particularly for students with lower prior academic achievement.

인공지능을 활용한 적응형 학습은 최근 국내 대학들이 직면하고 있는 학생들의 기초학력 저하와 학습격차 증가 등의 문제해결을 위한 방편이 될 수 있다. 인공지능 기반 적응형 학습이 성공적으로 대학 수업에 도입되고 실천되기 위해서는 교수자의 적극적인 관심과 참여가 요구된다. 이에 본 연구에서는 대학 교수들을 대상으로 적응형 학습에 대한 인식을 분석하여 대학 수업에서의 적응형 학습 구현을 위한 방안을 제안하고자 하였다. 이를 위하여 수도권 소재 A대학 교수들을 대상으로 온라인 설문을 통해 자료를 수집하였으며, 162명의 교수들이 응답에 참여하였다. 설문 분석 결과 교수들은 학생 맞춤형 피드백 제공의 어려움, 학생들의 사전학습 부족 및 기초학력 저하를 수업 운영에서의 문제로 높게 인식하고 있었다. 또 적응형 학습에 대한 교수들의 지식 수준은 낮았지만, 적응형 학습 적용 의향은 높은 것으로 나타났 다. 적응형 학습 적용을 위한 지원방안으로는 활용이 쉽고 유용한 적응형 학습 시스템 제공에 대한 요구가 가장 높았다. 이러한 결과를 바탕으로 대학에서의 적응형 학습 적용의 가능성을 논의하고, 적응형 학습의 성공적 도입과 적용을 위한 구체적인 방안을 제언하였다.

This study aimed to analyze the level of professors’ understanding and perception of adaptive learning and proposed how college can implement successful adaptive learning in college classes. For research purposes, online survey was conducted by 162 professors of A university in capital region. As a result, professors seemed to feel pressure to provide students personalized feedback and gave concerned that students don’t study enough in advance before participating in class. It was also found that professors realized that they have low level of understanding about adaptive learning, while they revealed intention to make use of adaptive learning in their class. They also answered that adaptive learning system is the most helpful support for encouraging professors to apply adaptive learning in real class. We proposed what is required to encourage professor to implement adaptive learning in their class.

Metaverse, our virtual reality, is traversed via an Avatar linked to a user profile through Personal Identifiable Information (PII). To secure this PII from causing attacker infiltration, only authorised users should access these avatars permitted by the authentication systems. The authentication systems are researched to be resilient against attackers’ manipulations. These systems rely on dynamic and real-time sensor data rather than static information from the user for authentication. Dynamic sensor data captured through Head Mounted Displays (HMDs) is highly classifiable with Machine learning (ML) and Deep Learning (DL) algorithms. Over time, the model training requires an upgrade through evolution in data processing and learning. Self-learning —Adaptive learning can lead this system to transform with its learn-evolve-adapt learning strategy. Therefore, our study attempts to explore authentication systems developed for HMDs, capturing realtime dynamic sensor data. With its results, we concluded that these systems are highly sensitive while processing the sensor data. We list out the risk factors of utilising adaptive learning for an authentication system based on neurometric data combined with biometric data. This study will be the state of the art for the self-learning algorithms for biometric and neurometric data-based authentication systems.

본 연구는 인공지능디지털교과서(AIDT)를 기반으로 한 학생 맞춤형 사회정서학습(SEL) 증진 시스템을 설계하 였다. 학습 분석, 정서 분석, 맞춤형 피드백 기능을 통합하여 개념적 모델과 시스템 아키텍처를 개발하고, 전문가 대상 설문조사와 인터뷰, 패널 토론을 통해 설계 타당성을 검토하였다. Pearson 상관분석 결과, 각 핵심 요소 간의 유의미한 관계가 확인되었으며, 전문가들은 AIDT 기반 SEL 시스템이 개별 학습자의 특성을 반영한 실시간 맞춤 형 피드백 제공이 가능하다는 점에서 높은 설계 타당성을 지닌다고 평가하였다. 이러한 결과는 향후 해당 시스템 이 학습 성취와 정서적 발달을 동시에 지원할 수 있는 가능성을 보여준다. 향후 연구에서는 실제 교육 현장에서의 적용을 통한 실증적 효과 검증과 함께, AI 기반 정서 분석 기술의 신뢰성 확보를 위한 추가 연구가 필요하다.

This study designed a student-personalized Social-Emotional Learning (SEL) enhancement system based on the Artificial Intelligence Digital Textbook (AIDT). By integrating learning analytics, emotion analysis, and personalized feedback, a conceptual model and system architecture were developed. Surveys, in-depth interviews, and panel discussions with experts were conducted to validate the system design. Pearson correlation analysis confirmed significant relationships among the system’s core components. Experts evaluated the system as having high design validity, particularly in delivering real-time personalized feedback tailored to individual learners. These findings suggest the potential of the system to support both academic achievement and socio-emotional development. Future research should focus on empirical validation through real-world implementation and improving the reliability of AI-based emotion analysis.

With the surge in mobile traffic and the demand for wide coverage, Low-Earth Orbit (LEO) satellite internet has garnered significant attention. However, the limited battery capacity of LEO satellites necessitates efficient energy management. Meanwhile, content caching presents a promising solution to reduce transmission delays and energy consumption of LEO satellites by caching popular files nearby end users. Unlike terrestrial networks, content caching in LEO satellites requires joint optimization with multi-hop routing due to inherent structural differences in ISL (Inter-Satellite Link) networks. In this paper, we formulate an optimization problem aimed at minimizing the average transmission power of LEO satellites while ensuring timely content delivery. To this end, we apply Lyapunov optimization framework to obtain a Hierarchical Caching and Routing (HCR) algorithm where content caching decisions are made over longer periods while routing decisions are made over shorter periods. Finally, via extensive simulations, we demonstrate that the HCR algorithm excels existing algorithms, showing up to an average of 54% and 67% reduction in average queue backlog and average power consumption, respectively.

Continual Test-Time Adaptation (CTTA) is essential for adapting models to target data in changing environments while retaining prior knowledge. However, previous methods overlook class imbalance, which limits performance on minor but essential objects. To address this, we propose AdaCoTTA, which applies confidence-guided adaptive learning to improve training stability. These mechanisms mitigate class imbalance, catastrophic forgetting, and error accumulation in CTTA. We evaluate AdaCoTTA on diverse CTTA benchmarks for semantic segmentation. It achieves state-of-the-art performance, improving average mIoU by 0.6% on ACDC and over 4% for minor classes, highlighting its effectiveness in scenarios. Code is available at?https://github.com/junghyeon0427/AdaCoTTA.

The channel state information (CSI) is essential for the base station (BS) to schedule user equipments (UEs) and efficiently manage the radio resources. Hence, the BS requests UEs to regularly feed back the CSI. However, frequent CSI reporting causes large signaling overhead. To reduce the feedback overhead, we propose two machine learning-based approaches to adjust the CSI feedback interval. We use a deep neural network and reinforcement learning (RL) to decide whether an UE feeds back the CSI. Simulation results show that the RL-based approach achieves the lowest mean squared error while reducing the number of CSI feedback transmissions.

To enable optical interconnect fluidity in next-generation data centers, we propose adaptive transmission based on machine learning in a wavelength-routing network. We consider programmable transmitters that can apply possible code rates to connections based on predicted bit error rate (BER) values. To classify the BER, we employ a preprocessing algorithm to feed the traffic data to a neural network classifier. We demonstrate the significance of our proposed preprocessing algorithm and the classifier performance for different values of and switch port count.

With the increasing sophistication of cyber threats, traditional Intrusion Detection Systems (IDS) often fail to adapt to evolving attack patterns, leading to high false positive rates and inadequate detection of zero-day attacks. This study proposes the Deep Q-Learning Intrusion Detection System (DQ-IDS), a novel reinforcement learning (RL)-based approach designed to dynamically learn network attack behaviors and continuously enhance detection performance. Unlike conventional machine learning (ML) and deep learning (DL)-based IDS models that depend on static, pre-trained classifiers, DQ-IDS employs Deep Q-Networks (DQN) with experience replay and adaptive ε-greedy exploration to autonomously classify benign and malicious network traffic. The integration of experience replay mitigates catastrophic forgetting, while adaptive exploration ensures an optimal trade-off between learning efficiency and threat detection. A reward-driven training mechanism reinforces correct classifications and penalizes errors, thereby reducing both false positive and false negative rates. Extensive empirical evaluations on real-world network datasets demonstrate that DQ-IDS achieves a detection accuracy of 97.18%, significantly outperforming conventional IDS solutions in both attack detection and computational efficiency. This work introduces a paradigm shift toward adaptive, self-learning cybersecurity systems capable of real-time, robust threat mitigation in dynamic network environments.

A distributed congestion control must be adaptable to varying target communication ranges as cellular V2X (C-V2X) is evolving to support flexible coverage suitable for various service scenarios. This study proposes range-adaptive distributed power control (Ra-DPC) based on deep reinforcement learning (DRL) with the Monte Carlo policy gradient algorithm. A key finding is that the agents learn Ra-DPC more effectively when the cumulative interference power of the subchannels is adopted as the state of the DRL model, rather than the channel busy ratio. The proposed Ra-DPC algorithm performs better in energy efficiency and packet delivery ratio than the existing technologies.

Large-scale pre-trained vision models such as ViT, CLIP, and SAM provide strong foundations for diverse vision tasks, motivating recent Mixture-of-Experts (MoE) approaches that combine multiple experts. However, existing methods often rely on static or implicit routing strategies, limiting adaptability to task semantics and input characteristics. We propose a task-adaptive vision expert routing framework based on competency learning guided by predictive uncertainty. We define expert competency as the relative reduction in predictive uncertainty induced by inter-expert interaction, and formulate expert routing as a learning problem driven by this signal. Our method uses task embeddings derived from textual descriptions to guide expert routing, refines expert features through cross-expert interaction, and aggregates them adaptively into a unified representation. By directly optimizing routing and feature composition using an uncertainty-based competency signal, the model learns how expert collaboration improves task-specific prediction reliability. Extensive experiments on diverse vision tasks demonstrate superior generalization performance and adaptive routing behavior aligned with task semantics.

This study introduces the Multi-Agent, Multi-Parameter, Interaction-Driven Contention Window Optimization (M2I-CWO) algorithm, a novel Multi-Agent Deep Reinforcement Learning (MADRL) framework designed to optimize multiple CW parameters in IEEE 802.11 Wireless LANs. Unlike single-parameter or specialized multi-agent methods, M2I-CWO employs a Dueling-DQN architecture and an Adaptive Interaction Reward Function?spanning independent, cooperative, competitive, and mixed modes?and accommodates Hierarchical Multi-Agent System (HMAS) or Federated RL (FRL) for further scalability. First, multiple CW parameters are simultaneously adjusted to enhance collision management. Second, M2I-CWO consistently achieves throughput improvements in both static and dynamic scenarios. Extensive results confirm M2I-CWO's superiority in efficiency and adaptability.

To manage the big data in constraint resources has difficulties and challenges. The power and the cost can be saved when the cloud services are used to process and store the data. However, the data includes the personal information that can be sensitive and should be hidden from the others. So we propose the privacy-preserving classification scheme for image data. The pixel-based learning is the scheme that is adapted to the cryptosystem, and is used to classify the encrypted images. Our proposed deep learning model has the convolutional layers that has the same size of the kernel with the block size in the cryptographic algorithm. The experiment results show that it can improve the accuracy on classification of encrypted images, and make it possible to use the private data securely.

This paper introduces a system that uses Functional Electrical Stimulation (FES) for finger flexion control aimed rehabilitation for stroke patients. To address the variability in electrode between patients, Reinforcement Learning is applied together with a switching network that allows automatic electrode selection. This results in an adaptable system that does not require rigorous searching of the patient’s optimal stimulation points. Data that supports the differences in the stimulation location for individuals as well as the ability of the system to converge automatically to a stimulation point is presented.

This study presents model-free reinforcement learn ing methods for economic and ecological adaptive cruise control (Eco-ACC) of connected and autonomous electric vehicles. For model-free optimal control of Eco-ACC, we applied two reinforcement learning methods, Deep Q-Network (DQN) and Deep Deterministic Policy Gradient (DDPG), in which deep neural networks of actors and critics were trained using IPG CarMaker simulations. For performance demonstrations, the HWFET, US06, and WLTP Class 3b driving cycles were used to simulate the front vehicle, and the energy consumptions of the host vehicle and front vehicle were compared. In high-fidelity IPG CarMaker simulations, the proposed reinforcement learning- based Eco-ACC methods demonstrated approximately 3–5% and 10–14% efficiency improvements in highway and city-highway driving scenarios, respectively, when compared with the front vehicle. A video of the CarMaker simulation is available at https://youtu.be/DIXzJxMVig8.

본 논문에서는 가시광 통신시스템의 조명과 통신의 기능을 동시에 사용할 때, 조명의 Dimming Level 향상과 Flicker 현상을 줄일 수 있는 기법에 대해서 제안한다. 가시광 통신은 통신과 조명의 성능을 함께 만족해야 한다. 그러나 기존의 Data Code Mapping 방식은 전체 조명의 밝기를 감소시키는 결과를 나타낸다. 이는 조명의 성능 저하와 Flicker 현상을 유발한다. 이를 해결하기 위해, 본 논문에서는 전송 알파벳에 대해서 binary code 할당하고, 문자열에 알파벳의 발생 빈도율에 따라 할당된 binary code를 최적화하여 매칭 시키는 적응 학습형 코드 할당 기법을 제안하였다. 이를 통해, 각각의 문자열의 최대 Dimming level을 유지하면서 동시에 ‘OFF' 패턴이 연속적으로 발생하지 않도록 코드를 할당하여 통신 기능뿐만 아니라 조명으 로써의 역할을 충실히 할 수 있는 기법에 대해서 연구하였다. 성능평가 결과, 전체 통신 성능에 큰 영향을 주지 않으면서, '1'의 발생 빈도가 유의미하게 증가하였고 반대로 연속적인 '0' 빈도율이 감소하여 시스템의 조명 성능이 크게 향상된 것을 보였다.

In this paper, when the lighting and communication functions of the visible light communication system are used at the same time, we propose a technique to reduce the dimming level and flicker of the lighting. Visible light communication must satisfy both communication and lighting performance. However, the existing data code method results in reducing the brightness of the entire lighting. This causes deterioration of lighting performance and flicker phenomenon. To solve this problem, in this paper, we propose an adaptive learning code allocation technique that allocates binary codes to transmitted characters and optimizes and matches the binary codes allocated according to the frequency of occurrence of alphabets in character strings. Through this, we studied a technique that can faithfully play the role of lighting as well as communication function by allocating codes so that the 'OFF' pattern does not occur continuously while maintaining the maximum dimming level of each character string. As a result of the performance evaluation, the frequency of occurrence of '1' increased significantly without significantly affecting the overall communication performance, and on the contrary, the frequency of consecutive '0' decreased, indicating that the lighting performance of the system was greatly improved.