- P-ISSN1738-6764
- E-ISSN2093-7504
- KCI
Vol.21 No.4
Abstract
Modern convolutional neural networks (CNNs) are essential in information and communication technology (ICT) applications, including edge computing, IoT devices, and mobile platforms, where energy efficiency and throughput are critical. These systems increasingly utilize multi-precision arithmetic to optimize accuracy and resource efficiency. However, traditional methods that assign separate fixed-precision multipliers for different bit-widths are inefficient, as the largest multiplier often dominates the critical path, limiting overall performance. In this paper, we introduce two scalable, power-efficient multiplier architectures with runtime reconfigurability: R4RC16 and R4RC32. These architectures are designed for CNN acceleration under multi-precision pruning. Each design features a low-power mode (8-bit) and a default mode (16-bit for R4RC16 and 32-bit for R4RC32), allowing for dynamic precision adjustment during inference with minimal overhead. When operating in low-power mode, our proposed multipliers achieve up to 7.6× greater energy efficiency compared to state-of-the-art approximate logarithmic multipliers, and up to 13.8× compared to approximate Booth-based designs. Additionally, they provide 2× (for 16-bit) and 4× (for 32-bit) higher throughput than exact 8-bit multipliers when processing pruned CNN workloads. Notably, the overhead in low-power mode is nearly independent of the full bit-width, resulting in a nearly constant power-delay product across both 16-bit and 32-bit designs. These findings highlight the significance of reconfigurable arithmetic units as critical components of ICT infrastructure that support healthcare, education, and multimedia, enabling CNNs to dynamically balance accuracy, energy, and throughput with less than 1% area overhead.
Abstract
The rapid expansion of digital transformation across industry, government, education, healthcare, and finance has positioned Artificial Intelligence (AI) as a central driving force of societal change. AI technologies are being integrated into diverse domains, restructuring industrial systems, innovating public services, and enhancing personalized quality of life. They significantly improve efficiency and productivity while creating new markets and employment opportunities. In particular, the rise of generative AI marks a qualitative shift in the technological paradigm, extending its influence into high-level human activities such as information creation, creative work, and decision support. However, these technological advancements and societal diffusion simultaneously generate a range of ethical and social risks. Algorithmic bias, opacity, privacy violations, workforce displacement, and unclear responsibility attribution undermine trust and societal acceptance of AI. Generative AI introduces additional concerns, including the veracity of generated content, copyright violations, and the proliferation of deepfakes, while philosophical and legal debates about accountability for AI-generated outcomes remain unresolved. In response to these challenges, this study re-examines the core principles and responsibility structures required for AI research ethics in the digital transformation era. It investigates an integrated research ethics framework grounded in three pillars: fairness, accountability, and human-centered design (HCD). Reflecting the expanding role of generative AI and evolving Human–Computer Interaction (HCI) dynamics, the study proposes a practical, operational evaluation model that overcomes the limitations of principle-centric ethical discourse. Specifically, it introduces the Integrated FAH–AIL Evaluation Framework (IFAEF), which combines the FAH-75 index for ethical compliance with the AIL-5 index that assesses AI intelligence and risk levels. As AI evolves from a tool to a “partner-like entity” influencing societal decision-making and human activity, mere technical performance is insufficient. Ethical frameworks must prioritize human dignity, societal context, and value-based judgment. Accordingly, AI research ethics must function not as a passive regulatory mechanism but as a strategic foundation for the sustainable coexistence of technology and society. This requires a multidimensional collaborative structure involving government, industry, academia, and civil society. This paper aims to contribute to this goal by examining the ethical conditions necessary throughout the AI development and deployment lifecycle, and by presenting actionable directions and quantitative evaluation standards for researcher ethics suited to the digital transformation era. In doing so, it offers a concrete foundation for strengthening the accountability, fairness, and human-centeredness of AI technologies.
Abstract
This work investigates the outage characteristics of a simultaneous wireless information and power transfer (SWIPT) opportunistic amplify-and-forward (OAF) relay network operating over Rayleigh fading channels. In the considered setup, each relay adopts a power-splitting (PS) protocol that divides the received radio-frequency signal for concurrent energy harvesting and data relaying. By formulating the partial channel state information (CSI)-based SWIPT OAF scheme as an equivalent generalized non-SWIPT relay model, tractable analytical expressions are derived to approximate the outage probabilities of both the indirect and combined transmission links, yielding reliable estimates of the true system behavior. Furthermore, an exact closed-form expression for the indirect link outage probability is obtained. Monte Carlo simulations confirm that the proposed analytical framework accurately captures the outage performance under a wide range of signal-to-noise ratio (SNR) conditions and relay selection scenarios. The developed analysis provides a convenient and insightful means for evaluating and optimizing SWIPT-assisted wireless sensor systems, thereby supporting their practical realization in energy-limited scenarios such as wireless sensor and Internet of Things (IoT) networks.
Abstract
This study explores how the choice between “Korea” and “South Korea” affects the readability of policy documents. An analysis of 89 documents found that those using “South Korea” had slightly higher average readability scores (M = 43.46) compared to those using “Korea” (M = 38.37). However, this difference was not statistically significant (t = 0.47, p = 0.64). This null finding suggests that the choice of country name may not significantly impact the readability of KIEP policy reports. This result is also reassuring, alleviating initial concerns that authors who ignore international naming conventions might produce less readable texts. Furthermore, this pilot study lays the groundwork for future regression analyses examining factors influencing readability, such as author gender, education, and the use of passive voice.
Abstract
This study explores the acceptance of art created by generative AI, focusing on cultural and artistic forms such as novels, paintings, and music. It proposes a new model for understanding the acceptance of generative AI art. The research questions examine how external factors, specifically presence and immersion, affect internal factors like perceived usefulness and ease of use, and how these factors influence user satisfaction and intention to continue using the technology. A survey was conducted, yielding 450 valid responses. The results revealed significant correlations among key variables: presence had a strong positive correlation with perceived usefulness (r = 0.651, p < .001), and immersion was also positively correlated with perceived usefulness (r = 0.644, p < .001). Furthermore, perceived usefulness was significantly linked to content satisfaction (r = 0.675, p < .001), while perceived ease of use showed a positive correlation with technological satisfaction (r = 0.618, p < .001). Additionally, the correlation between content satisfaction and intention to continue using the technology was high (r = 0.823, p < .001). This study is significant as it presents an initial theoretical model for generative AI art content and empirically confirms the characteristics, acceptance factors, and experiential effects of generative AI in the fields of art and culture.
Abstract
This study empirically examines follower gender as a crucial moderator connecting ethical leadership (EL) to organizational citizenship behavior (OCB) and, subsequently, to department performance. We analyzed survey data from 485 employees in Chinese SMEs using AMOS 18.0 and SPSS 18.0 to investigate both mediating and moderating effects, thus clarifying how EL influences performance. The results indicate that EL enhances department performance through OCB, with a stronger EL-OCB relationship observed among female followers compared to male followers. These findings provide practical guidance: promote ethical role modeling and procedural justice to encourage OCB, implement transparent decision-making and fair recognition to strengthen prosocial behavior, and customize communication, feedback, and recognition strategies in gender-responsive ways to optimize OCB’s impact on performance. By moving beyond a leader-centric perspective to consider follower characteristics, this study delineates when and how EL translates into unit-level performance via OCB.
Abstract
As the demand for immersive content that facilitates physical interactions with users continues to rise, so does the need for such content. The evolution of web technologies is expected to lead to a surge in 3D content that can be executed directly within web environments, providing a 3D experience across multiple platforms. The increasing availability of 3D content creation tools is likely to empower both professionals and novices to produce content more easily. Given the growing demand for 3D content, the trend towards cross-platform availability, and the simplification of content authoring tools, technologies that enable the production of interactive 3D content through Multi-Device Integration in web contexts are set to become essential. To address this demand, this paper presents an interaction API designed to streamline user interactions within web environments. By utilizing this API, developers can create interactive 3D content across various devices. The Interactive 3D Media Production framework enhances usability, accessibility, and platform independence by integrating diverse interaction modalities and providing plug-ins for other authoring tools.
Abstract
This study examines how Korean university students perceive English-language advertising for AI-based medical IT services. As AI-related medical technologies are increasingly promoted through English advertisements, it is necessary to understand how young consumers in a non-English-speaking context interpret and respond to such messages. This study employed Q-methodology to explore students’ subjective viewpoints. Twenty-three university students participated in a Q-sorting task using 16 statements related to English-language AI medical advertising. The analysis revealed three perception types: Informed Optimists, Critical Realists, and Trusting Pragmatists. The Informed Optimists regarded these advertisements as innovative and informative but were not easily influenced by emotional appeals. The Critical Realists showed a cautious and skeptical attitude, questioning the credibility and clarity of the advertising content. In contrast, the Trusting Pragmatists tended to accept the advertisements positively and preferred clear, factual, and professional messages. Across all types, English was generally associated with technological advancement, while concerns about comprehension and accessibility were also evident. These findings suggest the importance of audience-sensitive and trustworthy communication strategies in AI-based medical advertising.
Abstract
This study presents a four-tier integrated framework that combines Saussure's semiotics, Barthes' mythology, Hofstede's cultural dimensions, and Hall's encoding/decoding model to analyze cultural symbol exchange in cross-cultural cinema. By examining Crouching Tiger, Hidden Dragon (2000) and Mulan (2020), we demonstrate how this framework facilitates a systematic, multilayered analysis across structural, semantic, mythological, and reception dimensions. Through a detailed exploration of key symbols—the Green Destiny sword, bamboo forest, phoenix, and costume color systems—we uncover how cultural meanings are encoded within production contexts and interpreted differently by audiences based on their cultural backgrounds. Our findings indicate that symbols that engage all four theoretical dimensions create a richer cross-cultural resonance compared to those that prioritize visual spectacle over cultural authenticity. Methodologically, this study outlines how the framework's structured approach can be adapted for future large-scale content analysis and cross-platform reception studies, with potential applications in computational text analysis and multilingual audience research.
Abstract
In this study, we present a Retrieval-Augmented Generation (RAG)-based pipeline designed to extract key values from scientific literature on Silicon Carbide (SiC) crystal growth using the Physical Vapor Transport (PVT) method. To improve the relevance and completeness of the retrieved context, we implemented a hybrid retrieval strategy that combines dense retrieval via FAISS with sparse retrieval using BM25. We employed two distinct prompting approaches for key value extraction. The first approach addresses interactive user queries by utilizing the retrieved context to generate informed responses. The second approach, intended for bulk extraction, follows a two-step process: a binary classification prompt first checks for the presence of relevant information related to a query. If relevant information is confirmed, a subsequent prompt extracts the value under strict constraints—requiring exact phrasing without guessing or explanation. This binary pre-check significantly enhances the identification of true negative cases, thereby reducing irrelevant or missing data. For the generative component of our pipeline, we evaluated three large language models (LLMs): Llama 8B, Gemma 7B, and Mistral 7B, all operating on a local multi-GPU environment using FP16 precision. The results reveal differences in the efficiency of these models within our customized RAG system, particularly in their performance in extracting over 156 targeted technical key-value pairs from 13 benchmark papers.
Abstract
Current cross-country patent structural approaches show that comparative studies utilizing SAO (subject–action–object) structures are limited in specific technical domains. This paper analyzes the functional types of expressions in diabetes-related patent texts from the United States and China by examining verb usage and the semantic categories of subject and object phrases. It proposes a method to identify and classify patent functions based on SAO structures, aiming to uncover linguistic preferences and technological characteristics in functional expression across the two countries. The findings reveal significant differences between U.S. and Chinese diabetes patents regarding functional type, structural composition, and semantic expression style, indicating tendencies toward system integration-oriented and pharmacological mechanism-oriented expressions, respectively. Additionally, the analysis of time evolution and field distribution highlights distinct technical rhythms and layout characteristics in each country. The framework introduced in this paper offers methodological support and an empirical foundation for patent semantic modeling and cross-country technology comparison research.
Abstract
There have been various loss functions proposed to improve the training of neural networks with sigmoid activation output nodes. For neural networks with softmax activation output nodes, the cross-entropy loss function is commonly used for training and several attempts have been made to improve the performance of such networks by modifying the standard cross-entropy loss. However, rather than simply aiming to improve overall classification accuracy, it is often necessary to address misclassification costs differently depending on their real-world importance. In practice, the cost of errors can vary greatly across domains such as finance, security, insurance, and healthcare. From this perspective, this paper proposes a modified cross-entropy loss function designed to control the training of neural networks with softmax output nodes. The effectiveness of the proposed loss function is demonstrated through simulations on the CEDAR handwritten digit recognition task, showing that classification performance can be adjusted according to the order of the modified loss function. This approach will serve as the basis for designing a novel cost-sensitive learning method tailored to neural networks with softmax outputs.
Abstract
Recognizing the strategic importance of domain-specific high-performance computing (HPC), Korea designated six specialized national centers in 2023, with plans to expand this number to ten by 2030. To ensure effective operation and governance, a capability assessment providing a current and comparable view of these centers is essential. This study proposes a policy-linked framework for assessing the capabilities of Korea’s national HPC centers. The assessment model is organized into six categories: Foundation, Input, Process, Output, Outcome, and Policy Alignment, with a set of representative indicators specified for each category. These indicators were developed through an inductive analysis of relevant literature to identify key measures, along with a deductive integration of attributes specific to centers with statutory policy mandates. Each indicator is normalized against field-specific targets, and category-level and within-category weights are determined using the Analytic Hierarchy Process (AHP) to produce a single composite capability score for each center.
Abstract
Achievement emotions significantly influence students’ well-being, learning quality, and academic performance. This study addresses the overlooked achievement emotions in nursing by investigating their relationship with mindfulness and academic performance, with an emphasis on the mediating effects of achievement emotions. This study employed a cross-sectional design involving 203 nursing students in a university in C province. Participants responded to a survey designed to assess their level of mindfulness, achievement emotions, and academic performance. Structural relationships among mindfulness, achievement emotions, and academic performance were tested using data-driven and computational analyses, including structural equation modeling (SEM) with Mplus 8. In addition, path analysis with bootstrapping was conducted to validate the hypothesized pathways and to examine the mediating effects of achievement emotions. Mindfulness was positively related to positive achievement emotions and negatively related to negative achievement emotions. While positive achievement emotions enhanced academic performance, negative achievement emotions had an adverse effect. However, mindfulness did not directly influence academic performance. Notably, achievement emotions served as a complete mediator in the association between mindfulness and academic performance. The results suggest that mindfulness benefits nursing students’ academic performance by amplifying positive achievement emotions and diminishing negative ones. The mediation effect highlights the importance of achievement emotions in nursing education, highlighting the necessity of educational programs that enhance students’ emotional experiences. These findings can serve as a crucial foundation for designing initiatives fostering mindfulness, achievement emotions, and effective learning among nursing students.