Seok-Won Lee

In today’s rapidly evolving digital landscape, organizations confront increasingly sophisticated cyber threats with profound political, economic, and social implications. Among them, Advanced Persistent Threats (APTs) are particularly critical due to their persistence, strategic intent, and multi-stage lifecycles that compromise diverse assets while evading detection. Meanwhile, the emergence of cyber-physical-social (CPS) systems has created complex environments in which cyber, physical, and social domains are tightly interlinked. Traditional approaches—such as attack trees, attack patterns, cyber kill-chain models, and conventional risk assessment methods—remain limited, as they emphasize technology-level analysis and isolated countermeasures but fail to adequately capture multi-layered dependencies and adaptive APT behavior. To address these limitations, this study proposes a unified security risk assessment framework for CPS environments. The framework integrates four key components: (1) knowledge-based APT attack scenario generation and TTP-driven threat modeling; (2) construction of a Mission-Critical Target System (MCTS) comprising vertically and horizontally interconnected layers with extended criticality evaluation; (3) adaptive security requirements engineering to support continuous identification, specification, and refinement of defense measures; and (4) evidence-based risk assessment that incorporates technical, organizational, and social interdependencies to enable iterative, layered defense planning. Ontology-supported knowledge representation is employed to structure threats, assets, and security requirements and to improve traceability and auditability, while risk likelihoods are updated through evidence-driven evaluation rather than automated semantic inference. Validated through theoretical analysis and empirical case studies, the framework effectively captures the dynamic and multidimensional characteristics of APT behavior in CPS contexts and demonstrates improved comprehensiveness, precision, adaptability, and practical relevance compared with traditional assessment approaches.

Recent studies have shown that AI models can memorize specific data records, resulting in sensitive data exposure through model access. Current privacy-enhancing technologies often overlook the crucial, context-dependent nature of privacy risk as they largely fail to account for the inherent relationships and complex interactions between data records, leading to high risks associated with memorization and potential data aggregation. Our research first investigates two key factors influencing AI privacy risks: implicit connections and data redundancy. These experiments have shown that AI models learn subtle links between private data, even when they are discretely distributed. To address the privacy issue, we introduce PrivGraph, a hierarchically structured knowledge graph for modeling and aggregating private information. Based on PrivGraph, we introduce the Sensitivity Level Factor (SLF) to quantify the degree to which an individual’s private information is embedded in the data. In addition, we propose a PrivGraph-based knowledge probing method to facilitate post-training privacy assessments. Our experiments demonstrated that PrivGraph achieves comparable performance to existing models in the Personally Identifiable Information (PII) detection task, while effectively modeling the aggregation of private information even with lengthy texts and data obtained from multiple origins. Finally, we discuss PrivGraph’s integration into the AI engineering lifecycle for full-spectrum, full-lifecycle, and traceable privacy protection.

This study introduces a novel framework to enhance the counter-unmanned aerial system (C-UAS) effectiveness of Close-In Weapon Systems (CIWS) through reinforcement learning (RL). To address the challenges posed by the erratic maneuvering and swarm tactics of small UAS, we propose a hierarchical reinforcement learning (HRL) architecture integrated with a curriculum learning strategy. Leveraging a high-fidelity physics-based simulation, our approach employs a two-tiered decision-making structure that decouples high-level target selection from low-level fire control and weapon aiming. Experimental results demonstrate that the proposed RL-based agent significantly outperforms a sophisticated, model-based predictive controller designed to emulate the engagement logic of modern operational defense systems. The developed system exhibits robust performance across diverse engagement scenarios, including multi-drone attacks from various vectors, achieving quantifiable improvements in targeting precision and reaction time against highly agile threats. This research establishes a foundation for developing intelligent, autonomous defense systems capable of neutralizing emergent and evolving UAS threats in modern operational environments.

This research addresses the performance challenges of ontology-based context-aware and activity recognition techniques in complex environments and abnormal activities, and proposes an optimized ontology framework to improve recognition accuracy and computational efficiency. The method in this paper adopts the event sequence segmentation technique, combines location awareness with time interval reasoning, and improves human activity recognition through ontology reasoning. Compared with the existing methods, the framework performs better when dealing with uncertain data and complex scenes, and the experimental results show that its recognition accuracy is improved by 15.6% and processing time is reduced by 22.4%. In addition, it is found that with the increase of context complexity, the traditional ontology inference model has limitations in abnormal behavior recognition, especially in the case of high data redundancy, which tends to lead to a decrease in recognition accuracy. This study effectively mitigates this problem by optimizing the ontology matching algorithm and combining parallel computing and deep learning techniques to enhance the activity recognition capability in complex environments.

In embedded software for critical domains such as medical devices and defense, requirement traceability is essential for ensuring quality attributes. Standards and regulations mandate traceability between requirements and artifacts such as design elements and code. However, existing methods often overlook the hardware-dependent nature of embedded systems or conduct traceability retroactively, which may affect consistency. This study introduces a structured feature-based approach to component architecture design, bridging the gap between requirements and design to ensure traceability. The structured feature model supports traceability between functional requirements, software components, and hardware elements in embedded systems. A case study demonstrates that structured features can effectively map the requirements to design artifacts, helping to visualize relationships through a traceability matrix. Although the process is manual, structured features improve efficiency in the early stages of design and create traceable links between requirements and architectural elements.

Although sentiment analysis is pivotal to understanding user preferences, existing models face significant challenges in handling context-dependent sentiments, sarcasm, and nuanced emotions. This study addresses these challenges by integrating ontology-based methods with deep learning models, thereby enhancing sentiment analysis accuracy in complex domains such as film reviews and restaurant feedback. The framework comprises explicit topic recognition, followed by implicit topic identification to mitigate topic interference in subsequent sentiment analysis. In the context of sentiment analysis, we develop an expanded sentiment lexicon based on domain-specific corpora by leveraging techniques such as word-frequency analysis and word embedding. Furthermore, we introduce a sentiment recognition method based on both ontology-derived sentiment features and sentiment lexicons. We evaluate the performance of our system using a dataset of 10,500 restaurant reviews, focusing on sentiment classification accuracy. The incorporation of specialized lexicons and ontology structures enables the framework to discern subtle sentiment variations and context-specific expressions, thereby improving the overall sentiment-analysis performance. Experimental results demonstrate that the integration of ontology-based methods and deep learning models significantly improves sentiment analysis accuracy.

Images captured during marine engineering operations suffer from color distortion and low contrast. Underwater image enhancement helps to alleviate these problems. Many deep learning models can infer multi-source data, where images with different perspectives exist from multiple sources. To this end, we propose a multichannel deep convolutional neural network (MDCNN) linked to a VGG that can target multi-source (multi-domain) underwater image enhancement. The designed MDCNN feeds data from different domains into separate channels and implements parameters by linking VGGs, which improves the domain adaptation of the model. In addition, to optimize performance, multi-domain image perception loss functions, multilabel soft edge loss for specific image enhancement tasks, pixel-level loss, and external monitoring loss for edge sharpness preprocessing are proposed. These loss functions are set to effectively enhance the structural and textural similarity of underwater images. A series of qualitative and quantitative experiments demonstrate that our model is superior to the state-of-the-art Shallow UWnet in terms of UIQM, and the performance evaluation conducted on different datasets increased by 0.11 on average.

With the development of artificial intelligence technology, systems that can actively adapt to their surroundings and cooperate with other systems have become increasingly important. One of the most important factors to consider during the process of cooperation among systems is trust. Trust is a social concept that assumes that cooperation with an object will produce positive results in the direction we intend. Our objectives are to propose a method for defining trust during the requirements engineering phase in the process of developing self-adaptive systems and to define the trust evidence models required to evaluate the defined trust at runtime. To achieve this objective, we propose in this study a provenance-based trust-aware requirement engineering framework for self-adaptive systems. The framework helps system engineers derive the user’s requirements as a trust-aware goal model through analysis of the trust concept in the requirements engineering process. We also propose a provenance-based trust evidence model to evaluate trust and provide a method for defining this model for the target domain. Through the proposed framework, a system engineer can treat trust as a factor emerging from the requirements engineering phase for the self-adaptive system and understand the factors affecting trust using the standardized format.