-
- [Jun 30, 2025] seclab(Advisor: Prof. Hyoungshick Kim) Selected for Oral Session at ACL 2025
- The research conducted by seclab (Advisor: Prof. Hyoungshick Kim), in collaboration with the National Security Research Institute and KAIST, has been accepted to the main conference of the 2025 Annual Meeting of the Association for Computational Linguistics (ACL 2025), one of the most prestigious international conferences in the field of artificial intelligence and natural language processing. Notably, the paper was selected for the Oral Presentation session, which is reserved for only the top 243 papers out of more than 3,000 submissions. The paper was authored by Woo-Young Ko, a Ph.D. candidate at KAIST and a researcher at the National Security Research Institute, as the first author. Prof. Hyoungshick Kim participated as a co-researcher. XDAC: XAI-Driven Detection and Attribution of LLM-Generated News Comments in Korean The paper introduces XDAC, the world’s first explainable AI (XAI)-based detection framework that not only detects Korean news comments generated by large language models (LLMs) but also identifies the AI model that generated them. The system achieves an impressive 98.5% precision in detecting AI-generated comments, and it can also identify which LLM was used to generate the content with 84.3% accuracy—surpassing current state-of-the-art (SOTA) technologies. This research is highly valued for its contribution in the context of growing concerns about the potential for AI-driven opinion manipulation. It provides an effective defense mechanism that works well even in short, informal Korean comments, which are typically challenging to analyze. As a result, the research has garnered significant attention for its practical application, with potential for adoption by portal platforms and government institutions in the future. Project Source Code: https://github.com/airobotlab/XDAC/tree/main News Article: https://www.ddaily.co.kr/page/view/2025062717211207639
-
- 작성일 2025-07-02
- 조회수 220
-
- [Jun 26, 2025] Professor Kwang-Soo Kim AAI Lab, Approves Two Papers for Publication at ICML 2025
- AAI Lab's (Advisor: Kwang-Soo Kim) paper was accepted for publication in ICML (International Conference on Machine Learning) 2025, an excellent academic conference in the field of artificial intelligence. The first paper, "One-Step Generalization Ratio Guided Optimization for Domain Generalization", was written by Soo-Min Cho (Master's course) and Dong-Won Kim (Master's course). This paper was selected as a Spotlight Poster and was also selected as an oral presentation paper, which is less than 1% of all published papers. This paper proposes GENIE (Generalization-ENhancing Iterative Equalizer), a new optimization technique for solving the Domain Generalization (DG) problem. Existing DG methods have the risk of reinforcing spurious correlations that overfit a specific domain, and in particular, they have overlooked the problem that unbalanced updates between parameters hinder generalization performance. In this study, we introduce a new preconditioning-based optimization that induces parameter-specific update balance based on the One-Step Generalization Ratio (OSGR), an indicator that quantifies the degree to which each parameter contributes to generalization. GENIE measures OSGR in real time and dynamically assigns preconditioning coefficients to each parameter accordingly to prevent a small number of parameters from excessively driving learning. Additionally, noise injection and random masking ensure the stability and explorability of learning and prevent overfitting. Summary of Contributions 1. Proposal of GENIE Optimizer - Introduce OSGR-based preconditioning to prevent a small number of parameters from excessively driving updates and induce all parameters to contribute to generalization in a balanced manner. 2. Securing theoretical justification - Through OSGR-based analysis, we theoretically prove that GENIE balances the generalization contributions among parameters. - Theoretically prove that GENIE's preconditioning can make the generalization bound tighter through PAC-Bayes analysis. - Compared to existing optimization algorithms, GENIE achieves higher OSGR and mathematically outperforms in generalization performance. - Prove that it maintains the same convergence rate as SGD even in non-convex environments. 3. Excellent experimental results - Average 2~6% accuracy improvement over existing optimization techniques in 5 DG datasets (PACS, VLCS, OfficeHome, TerraIncognita, DomainNet). - Performance improvement confirmed when used as an optimizer for existing DG algorithms. - Excellent performance even in Single-Domain Generalization (SDG) environments. - Effectiveness proven through various analysis experiments. Abstract Domain Generalization (DG) aims to train models that generalize to unseen target domains but often overfit to domain-specific features, known as undesired correlations. Gradient-based DG methods typically guide gradients in a dominant direction but often inadvertently reinforce spurious correlations. Recent work has employed dropout to regularize overconfident parameters but has not explicitly adjusted gradient alignment or ensured balanced parameter updates. We propose GENIE (Generalization-ENhancing Iterative Equalizer), a novel optimizer that leverages the One-Step Generalization Ratio (OSGR) to quantify each parameter’s contribution to loss reduction and assess gradient alignment. By dynamically equalizing OSGR via a preconditioning factor, GENIE prevents a small subset of parameters from dominating optimization, thereby promoting domain- invariant feature learning. Theoretically, GENIE balances convergence contribution and gradient alignment among parameters, achieving higher OSGR while retaining SGD’s convergence rate. Empirically, it outperforms existing optimizers and enhances performance when integrated with various DG and single-DG methods. The second paper, "Federated Learning for Feature Generalization with Convex Constraints", was written by Dongwon Kim (Master's course), Donghee Kim (PhD course), and Seongguk Shin (PhD course). This paper was selected as a Poster presentation paper. In federated learning, the generalization performance degradation due to data heterogeneity between clients is a common problem. Local models are prone to overfitting to their own data distributions, and even generalizable features already acquired by local models can be distorted during the aggregation process. To solve this problem, we propose FedCONT. This method adaptively adjusts the update size according to the parameter strength of the global model, thereby preventing overemphasizing sufficiently learned parameters and enhancing parameters with insufficient learning. Specifically, FedCONST secures learning stability by utilizing linear convex constraints and maintains the generalization ability acquired by local models during the aggregation process. In addition, we demonstrate that FedCONST effectively improves the transferability and robustness of feature values through Gradient Signal-to-Noise Ratio (GSNR) analysis. As a result, FedCONST effectively aligns the optimal values of local and global models to mitigate overfitting and achieves stronger generalization performance across various FL environments, achieving state-of-the-art performance. Summary of Contributions 1. Proposal of FedCONST Method To simultaneously improve the generalization performance of client and server models in federated learning, we introduce convex constraints based on the magnitude of global parameters. This suppresses over-learned parameters and enhances under-learned parameters, thereby inducing balanced learning. 2. Securing Theoretical Justification We quantify the generalization contribution of each parameter through Gradient Signal-to-Noise Ratio (GSNR) analysis, and mathematically prove that the proposed convex constraints improve these contributions in a balanced manner. 3. Excellent experimental performance It shows excellent results in overall accuracy and generalization performance compared to existing methods such as FedAvg, FedProx, and FedDyn in various federated learning environments. 4. High scalability and compatibility The proposed method is independent of model architecture or existing algorithms, and can be easily combined with various federated learning optimization techniques to improve performance. Abstract Federated learning (FL) often struggles with generalization due to heterogeneous client data. Local models are prone to overfitting their local data distributions, and even transferable features can be distorted during aggregation. To address these challenges, we propose FedCONST, an approach that adaptively modulates update magnitudes based on the global model’s parameter strength. This prevents over-emphasizing well-learned parameters while reinforcing underdeveloped ones. Specifically, FedCONST employs linear convex constraints to ensure training stability and preserve locally learned generalization capabilities during aggregation. A Gradient Signal-to-Noise Ratio (GSNR) analysis further validates FedCONST's effectiveness in enhancing feature transferability and robustness. As a result, FedCONST effectively aligns local and global objectives, mitigating overfitting and promoting stronger generalization across diverse FL environments, achieving state-of-the-art performance.
-
- 작성일 2025-06-30
- 조회수 265
-
- [May 27, 2025] Professor Lee Ji-hyung Information and Systems Lab, 1 paper accepted for publication in ACL 2025
- One paper from the Information and Intelligence Systems Lab (Advisor: Lee Ji-hyung) has been accepted for publication at ACL 2025 (“The 63rd Annual Meeting of the Association for Computational Linguistics”), the premier international academic conference in the field of natural language processing. The paper is scheduled to be presented in Vienna, Austria in July. Title: “DCG-SQL: Enhancing In-Context Learning for Text-to-SQL with Deep Contextual Schema Link Graph” Authors: Jihyung Lee* (Master’s Program in Artificial Intelligence), Jinseop Lee* (Integrated Master’s and Doctoral Program in Artificial Intelligence), Jaehoon Lee (Master’s Program in Artificial Intelligence), Yunseok Choi (Assistant Professor, Department of Computer Education, Sungkyunkwan University/PhD in Software) (*Co-first author) In the paper “DCG-SQL: Enhancing In-Context Learning for Text-to-SQL with Deep Contextual Schema Link Graph,” we propose an example selection technique for effective In-context Learning in the Text-to-SQL task of converting natural language sentences into SQL queries. In-context Learning is a method that induces a model to solve a problem by utilizing a small number of examples, but existing example selection methods do not have a significant difference in performance compared to random selection, and especially, they show limitations in that the performance is greatly reduced in small language models such as Llama 3.1-8B. This suggests that existing methods rely on the intrinsic capabilities of the model itself rather than effectively utilizing in-context learning. To address this issue, this study proposes a method to select examples more suitable for in-context learning by utilizing a context-based schema connection graph that reflects the core information and contextual relationships between questions and database schemas. Through experiments on various Text-to-SQL benchmark datasets, we were able to confirm consistent performance improvements and practical effects regardless of the model size by improving SQL generation performance not only for hyper-scaled language models but also for small models. Abstract: Text-to-SQL, which translates a natural language question into an SQL query, has advanced with in-context learning of Large Language Models (LLMs). However, existing methods show little improvement in performance compared to randomly chosen demonstrations, and significant performance drops when smaller LLMs (e.g., Llama 3.1-8B) are used. This indicates that these methods heavily rely on the intrinsic capabilities of hyper-scaled LLMs, rather than effectively retrieving useful demonstrations. In this paper, we propose a novel approach for effectively retrieving demonstrations and generating SQL queries. We construct a Deep Contextual Schema Link Graph, which contains key information and semantic relationship between a question and its database schema items. This graph-based structure enables effective representation of Text-to-SQL samples and retrieval of useful demonstrations for in-context learning. Experimental results on the Spider benchmark demonstrate the effectiveness of our approach, showing consistent improvements in SQL generation performance and efficiency across both hyper-scaled LLMs and small LLMs.
-
- 작성일 2025-05-27
- 조회수 454
-
- [Apr 30, 2025] Professor Jongwook Lee Data Intelligence and Learning Lab Published 3 SIGIR 2025 Papers
- The Data Intelligence and Learning (DIAL, Professor: Jong-Wook Lee) lab has had three papers accepted for publication at SIGIR 2025, the world's most prestigious information retrieval conference, and will present them in Padua, Italy in July. [List of Papers] 1. Why is Normalization Necessary for Linear Recommenders? (SIGIR'25) 2. Linear Item-Item Models with Neural Knowledge for Session-based Recommendation (SIGIR'25) 3. DIFF: Dual Side-Information Filtering and Fusion for Sequential Recommendation (SIGIR'25) Study 1: Seongmin Park, Mincheol Yoon, Hye-young Kim, Jongwuk Lee, “Why is Normalization Necessary for Linear Recommenders?”, The 48th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), 2025 This study focuses on the fact that linear autoencoder (LAE)-based recommendation systems have comparable recommendation performance and fast inference speed to neural network-based models despite their simple structure. However, LAE faces two structural limitations: popularity bias, which over-recommends popular items, and neighborhood bias, which relies excessively on local correlations between items. To address these issues, in this paper, we propose a novel normalization method, Data-Adaptive Normalization (DAN), that can be applied to the LAE model. DAN is a normalization technique designed to flexibly control two biases depending on the characteristics of the data, and consists of two core components: (i) item-adaptive normalization and (ii) user-adaptive normalization. First, item-adaptive normalization controls the influence of popular items through the regularization parameter α and provides a denoising effect to LAE. This allows LAE to significantly improve the recommendation performance for unpopular items (tail items) by moving away from the performance centered on popular items (i.e., popularity bias) that the existing LAE mainly recommends. Second, user-adaptive normalization controls the neighborhood bias using the parameter β. This technique suppresses high-frequency components and preserves important low-frequency components, thereby helping to better reflect the overall global pattern rather than local correlations. The effectiveness of DAN is experimentally verified on six representative recommendation datasets (ML-20M, Netflix, MSD, Gowalla, Yelp2018, Amazon-book). LAE models (LAE_DAN, EASE_DAN, RLAE_DAN) applied with DAN showed consistent performance improvement over the existing LAE model on all datasets, and recorded performance improvements of up to 128.57% and 12.36% in tail items and unbiased evaluation, respectively. DAN also showed superior performance compared to the state-of-the-art collaborative filtering models. In addition, Figure 1 (Case study) shows the recommendation results for a specific user according to the regularization method, and the following observations were made: (1) LAE without regularization (W/O) recommends only five highly popular action movies even though the user watched three romantic movies. On the other hand, the three regularization methods (RW, Sym, DAN) effectively reflect user preferences by recommending “Step Up 2” as the top item related to “Step Up 1” viewed by the user. (2) DAN provides the most balanced recommendation that maintains user preferences while appropriately mitigating popularity bias. RW regularization still has strong popularity bias with 4 out of 5 items being popular. Sym regularization overly mitigates popularity bias with 4 out of 5 items being unpopular. DAN recommends the items most relevant to user preferences while balancing popular and unpopular items. Figure 1: Interaction history of user #91935 on the ML-20M dataset and the top-5 recommendation list of the four regularization methods. The red border is the head (top 20%) item, and the blue border is the tail (bottom 80%) item. Furthermore, this study analyzes how the effect of the regularization parameters (α, β) varies depending on the Gini index and homophily characteristics of the dataset, and also presents guidelines for setting parameters suitable for each dataset. Through this, it is shown that the proposed DAN technique can be established as a general and practical solution that can precisely control bias depending on the data characteristics. For more information about this paper, please refer to the following address. https://dial.skku.edu/blog/2025_dan Study 2: Minjin Choi, Sunkyung Lee, Seongmin Park, Jongwuk Lee, “Linear Item-Item Models with Neural Knowledge for Session-based Recommendation”, The 48th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), 2025 This study focuses on the problem of session-based recommendation (SBR), which predicts the next action based on the user’s current session interaction. The SBR field is largely divided into two paradigms. One is a neural network-based model that is strong in capturing complex sequential transition patterns, and the other is a linear item-item model that effectively learns co-occurrence patterns between items and provides fast inference speed. However, each paradigm is specialized in capturing different types of item relationships, and an effective integrated method to simultaneously achieve high accuracy of neural network models and efficiency of linear models is still lacking. Against this backdrop, in this paper, we propose a novel SBR model, LINK (Linear Item-Item model with Neural Knowledge), which effectively integrates knowledge from linear models and neural network models. LINK aims to achieve both high accuracy and fast inference speed by combining two types of knowledge within a single unified linear framework. To this end, LINK includes two core components. (i) LIS (Linear knowledge-enhanced Item-item Similarity model) enhances the linear model’s ability to capture item similarity (co-occurrence) and learns high-dimensional relationships between sessions through self-distillation technique. (ii) NIT (Neural knowledge-enhanced Item-item Transition model) effectively injects neural network knowledge into linear models through a unique method that distills complex sequential transfer knowledge from pre-trained arbitrary neural network models and utilizes it as a regularization term when learning linear models. As shown in Figure 2, the effectiveness of the LINK model has been verified through extensive experiments using six real-world SBR datasets, including Diginetica, Retailrocket, and Yoochoose. The experimental results show that LINK consistently and significantly improves performance (up to 14.78% in Recall@20 and up to 11.04% in MRR@20) compared to existing state-of-the-art linear SBR models (such as SLIST and SWalk) on all datasets. This demonstrates that LINK successfully overcomes the limitations of linear models by incorporating neural network knowledge. In addition, LINK maintains the high inference efficiency (up to 813 times fewer FLOPs), which is a key advantage of linear models, while showing competitive or superior prediction accuracy compared to complex state-of-the-art neural network models. Further analysis shows that linear models are strong in the relationship between unpopular items, while neural network models are strong in the complex patterns of popular items, and LINK effectively combines these two strengths to achieve balanced performance. Figure 2: Comparison of Accuracy (Recall@20) and Inference Operations (FLOPs) In conclusion, LINK presents a novel hybrid approach that provides a practical trade-off between accuracy and efficiency in the SBR field. In particular, the NIT component provides the flexibility to leverage knowledge from various models without being bound to a specific neural network architecture, making it a practical solution that can continuously improve performance as neural network models evolve in the future. For more information about this paper, please refer to the following address: https://dial.skku.edu/blog/2025_link Study 3: Hye-young Kim, Minjin Choi, Sunkyung Lee, Ilwoong Baek, Jongwuk Lee, “DIFF: Dual Side-Information Filtering and Fusion for Sequential Recommendation”, The 48th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), 2025 This study proposes a side-information integrated sequential recommendation (SISR) model that utilizes additional information such as category and brand in sequential recommendation, which predicts the next preferred item based on the user’s past consumption history. The proposed model, Dual Side-Information Filtering and Fusion (DIFF), removes noise in the user sequence and effectively fuses various attribute information to achieve more precise and expressive user preference modeling. DIFF includes the following three core techniques: Figure 3: Frequency signals and fusion techniques of sequential recommendation systems that integrate additional information (1) Frequency-based noise filtering: DIFF performs frequency domain transformation to remove signals that are not related to actual user preferences, such as accidental clicks or short-term interests. After converting the item ID and each attribute sequence to the frequency domain, it removes irregular or low-importance frequency components. This allows us to strengthen only the core signals that reflect actual user preferences, and enables more sophisticated noise removal by applying filtering to multiple sequences. (2) Dual multi-sequence fusion: DIFF utilizes intermediate fusion and early fusion methods, which have different strengths, together to effectively integrate the denoised sequences. We note that previous studies have tended to limit or exclude the use of early fusion methods due to concerns about information invasion, which has led to the overlooking of the ability to model correlations between various attributes. DIFF learns sophisticated user representations that encompass both IDs and attributes by integrating multidimensional attribute information through early fusion and supplementing ID-centric preference learning through intermediate fusion. Through the complementary combination of the two fusion methods, DIFF can effectively capture not only the overall structure of user tastes but also detailed attribute preferences. (3) Representation alignment to prevent information invasion: Item IDs and each attribute embedding are located in different representation spaces. Therefore, early fusion that combines them with a simple fusion function (e.g. summation, concatenation, gating) may cause information invasion, where specific information is overly emphasized or distorted. To prevent this, DIFF designs an alignment loss to make the vector spaces of item IDs and attribute embeddings close together, thereby maintaining appropriate differences while sharing meaning. DIFF has been validated on four representative public benchmark datasets (Yelp, Beauty, Toys, Sports), and has achieved superior performance in all indicators compared to existing state-of-the-art sequential recommendation models. In particular, it has demonstrated a new state-of-the-art performance by recording performance improvements of up to 14.1% and 12.5% on Recall@20 and NDCG@20, respectively. In addition, DIFF's robustness against noise is very remarkable. Considering noise in a realistic usage environment such as accidental clicks and temporary changes in interest in the user sequence, we conducted noise simulation experiments by randomly replacing items in the test sequence. As a result, DIFF showed the least performance degradation compared to other models even under low noise conditions of 5%, and maintained stable high performance even under high noise conditions of 25%. For more information about this paper, please refer to the following address: https://dial.skku.edu/blog/2025_diff
-
- 작성일 2025-04-30
- 조회수 680
-
- [Apr 30, 2025] Professor Lee Ji-hyung elected as new president of the Korean Society for Artificial Intelligence
- Professor Lee Ji-hyung Elected as New President of the Korean Society for Artificial Intelligence Professor Lee Ji-hyung of the Department of Software Elected as the 5th President of the Korean Society for Artificial Intelligence on April 16. Professor Lee has been serving as a director of the society and has contributed to the academic and technological development of the artificial intelligence field. In 2022, he served as the organizing committee chairman of the ‘Korean Society for Artificial Intelligence & Naver Fall Joint Academic Conference’ and played a major role in revitalizing academic exchange. Professor Lee Ji-hyung received the Sungkyunkwan Family Award in the Educational Achievement category from our school in 2019, and in 2022, he received the Byeon Jeong-nam Academic Award from the Korean Society for Intelligent Systems in recognition of his contributions to the development of artificial intelligence. In 2023, he received the Minister of Science and ICT Award in recognition of his contributions to fostering artificial intelligence talent, and continues to make outstanding achievements in all areas of education and research. Meanwhile, the Korean Artificial Intelligence Society, established in 2016, is actively promoting academic exchanges and industry-academia-research cooperation activities for the purpose of research and education in artificial intelligence and related fields such as computer vision, pattern recognition, natural language processing, bioinformatics, brain cognitive computing, and machine learning. Through this, it is playing a pivotal role in the development and spread of domestic artificial intelligence technology.
-
- 작성일 2025-04-30
- 조회수 662
-
- [Mar 28, 2025] Professor Kim Hyoungshick Security Engineering Laboratory (seclab) – Paper Accepted for Publication at...
- Security Engineering Laboratory (seclab) at SKKU (Advisor: Kim Hyoung-shick, https://seclab.skku.edu) – "Open Sesame! On the Security and Memorability of Verbal Passwords" Accepted for IEEE Symposium on Security and Privacy (S&P) 2025 The paper "Open Sesame! On the Security and Memorability of Verbal Passwords," conducted by Ph.D. candidate Kim Eun-soo and Professor Kim Hyoung-shick at the Security Engineering Laboratory, has been accepted for publication at the IEEE Symposium on Security and Privacy (S&P) 2025, one of the most prestigious conferences in the field of computer security. The study was conducted in collaboration with Professor Kim Doo-won of the University of Tennessee and alumnus Lee Ki-ho from the Security Engineering Laboratory (currently at ETRI). The research quantitatively analyzed the security and memorability of verbal passwords through two large-scale user experiments, demonstrating that verbal passwords offer a practical and secure alternative to traditional text-based passwords by overcoming their inherent limitations. In the first user experiment, verbal passwords freely generated by 2,085 participants were evaluated for both short-term and long-term memorability as well as security. Security testing conducted using the PassphraseGPT model—trained on over 20 million common English phrases—revealed that approximately 39.76% of the user-generated verbal passwords could be predicted within one billion guess attempts. In the second experiment, involving 600 participants, a password creation policy that enforced a minimum word count and incorporated a blocklist was implemented. This approach significantly improved security while maintaining ease of memorability. In long-term memory tests, 65.6% of users in the verbal password group were able to successfully recall their passwords, compared to 54.11% for text-based passwords. Moreover, the proportion of verbal passwords susceptible to guessing attacks was lower than that of text passwords, indicating a stronger resistance to such attacks. This research has been highly acclaimed for demonstrating that verbal passwords provide a practical and secure alternative to text-based passwords in scenarios where keyboard input is either impossible or inconvenient—such as with smart assistants, wearable devices, in-vehicle systems, and VR/AR environments. The study will be presented in May 2025 in San Francisco, California, USA. Abstract Despite extensive research on text passwords, the security and memorability of verbal passwords—spoken rather than typed—remain underexplored. Verbal passwords hold significant potential for scenarios where keyboard input is impractical (e.g., smart speakers, wearables, vehicles) or users have motor impairments that make typing difficult. Through two large-scale user studies, we assessed the viability of verbal passwords. In our first study (N = 2,085), freely chosen verbal passwords were found to have a limited guessing space, with 39.76% cracked within 10^9 guesses. However, in our second study (n = 600), applying word count and blocklist policies for verbal password creation significantly enhanced verbal password performance, achieving better memorability and security than traditional text passwords. Specifically, 65.6% of verbal password users (under the password creation policy using minimum word counts and a blocklist) successfully recalled their passwords in long-term tests, compared to 54.11% for text passwords. Additionally, verbal passwords with enforced policies exhibited a lower crack rate (6.5%) than text passwords (10.3%). These findings highlight verbal passwords as a practical and secure alternative for contexts where text passwords are infeasible, offering strong memorability with robust resistance to guessing attacks.
-
- 작성일 2025-03-28
- 조회수 1006
-
- [Mar 26, 2025] Professor Kim Hyoung-shick Security Engineering Lab – Two Papers Accepted for Oral Sessions at The Web..
- The Security Engineering Laboratory, under the supervision of Professor Kim Hyoung-shick, in collaboration with Professor Kim Doo-won from the University of Tennessee, has had two research papers accepted for oral sessions at The Web Conference (WWW) 2025, one of the premier international conferences in the web domain. In this research, alumnus Lee Ki-ho, a former member of the Security Engineering Laboratory (currently at ETRI), participated as a visiting researcher at the University of Tennessee and collaborated with Professor Kim Hyoung-shick. Both papers, based on extensive empirical data, quantitatively analyze the characteristics and structures of phishing attacks. They have been highly acclaimed for providing a fundamental understanding of phishing attacks and proposing new countermeasures. The presentations are scheduled to take place in May 2025 in Sydney, Australia. Paper 1. 7 Days Later: Analyzing Phishing-Site Lifespan After Detected This paper presents an empirical study analyzing the lifetime and evolution of phishing sites after detection. Over a period of five months, 286,237 phishing URLs were tracked at 30-minute intervals to examine the attack patterns of phishing sites, shedding light on why the effectiveness of conventional phishing detection strategies is diminishing. Phishing sites have a short lifespan—with an average survival time of 54 hours and a median of 5.46 hours—highlighting the limitations of training and detection approaches. For instance, Google Safe Browsing detects phishing sites, on average, 4.5 days after their emergence; however, 84% of phishing sites cease operations before detection, demonstrating the inherent limitations of such detection methods. Paper 2. What's in Phishers: A Longitudinal Study of Security Configurations in Phishing Websites and Kits This paper presents a systematic analysis of phishing infrastructure by comprehensively examining the security configurations and structural vulnerabilities based on a combined dataset of 906,731 phishing websites and 13,344 phishing kits collected over a period of 2 years and 7 months. The study has attracted attention for proposing a proactive strategy that leverages the structural weaknesses of phishing sites to neutralize the attack infrastructure, thereby moving away from traditional passive detection and blocking methods and towards an early shutdown approach for phishing sites.
-
- 작성일 2025-03-28
- 조회수 1093
-
- [Mar 17, 2025] Professor Lee Ho-jun Research Laboratory (SSLab) IEEE S&P 2025 Paper Acceptance Announcement
- [IEEE S&P 2025 Acceptance Announcement – SSLab, Professor Hojoon Lee] The paper from the System Security Laboratory (SSLab), under the supervision of Professor Hojoon Lee, has been accepted for publication at IEEE S&P 2025, one of the four premier international conferences in the security field. The paper is scheduled for presentation in May in San Francisco, California, USA. Title: IncognitOS: A Practical Unikernel Design for Full-System Obfuscation in Confidential Virtual Machines Authors: Kha Dinh Duy, Jaeyoon Kim, Hajeong Lim, Hojoon Lee Summary: Recent works have repeatedly proven the practicality of side-channel attacks in undermining the confidentiality guarantees of Trusted Execution Environments such as Intel SGX. Meanwhile, the trusted execution in the cloud is witnessing a trend shift towards confidential virtual machines (CVMs). Unfortunately, several side-channel attacks have survived the shift and are feasible even for CVMs, along with the new attacks discovered on the CVM architectures. Previous works have explored defensive measures for securing userspace enclaves (i.e., Intel SGX) against side-channel attacks. However, the design space for a CVM-based obfuscation execution engine is largely unexplored. This paper proposes a unikernel design named IncognitOS to provide full-system obfuscation for CVM-based cloud workloads. IncognitOS fully embraces unikernel principles such as minimized TCB and direct hardware access to render full-system obfuscation feasible. IncognitOS retrofits two key OS components, the scheduler and memory management, to implement a novel adaptive obfuscation scheme. IncognitOS's scheduling is designed to be self-sovereign from the timer interrupts from the untrusted hypervisor with its synchronous tick delivery. This allows IncognitOS to reliably monitor the frequency of the hypervisor's possession of execution control (i.e., VMExits) and adjust the frequency of memory rerandomization performed by the paging subsystem, which transparently performs memory rerandomization through direct MMU access. The resulting IncognitOS design makes a case for self-obfuscating unikernel as a secure CVM deployment strategy while further advancing the obfuscation technique compared to previous works. Evaluation results demonstrate IncognitOS's resilience against CVM attacks and show that its adaptive obfuscation scheme enables practical performance for real-world programs.
-
- 작성일 2025-03-28
- 조회수 1082
-
- [Mar 4, 2025] Professor Simon S Woo DASH Lab One paper accepted at EuroS&P 2025 from
- The Data-driven AI & Security HCI Lab (DASH Lab, Advisor: Simon S. Woo) has had one System of Knowledge (SoK) paper accepted for publication at the 10th IEEE European Symposium on Security and Privacy (Euro S&P), a prestigious international conference covers Machine Learning Security, System & Network Security, Cryptographic Protocols, Data Privacy. The papers will be presented in July in Venice, Italy. SoK: Systematization and Benchmarking of Deepfake Detectors in a Unified Framework, EuroS&P 2025 Authors: Binh Le and Jiwon Kim (Ph.D. Student, Department of Software), Simon S. Woo (Professor, Department of Software, Sungkyunkwan University) This work is jointly performed with CSIRO Data61 as an international collaboration. Paper Link: https://arxiv.org/abs/2401.04364
-
- 작성일 2025-03-04
- 조회수 1041
-
- [Mar 4, 2025] 2024 SKKU Fellowship 10 Professors Selected(Professor Hyungsik Kim of College of Software Convergence)
- 2024 SKKU Fellowship 10 Professors Selected Our university has selected Professor Minwoo Kim of the College of Social Sciences, Professor Ahyoung Seo of the College of Business Administration, Professor Donghee Son of the College of Information and Communications, Professor Jaehyuk Choi of the College of Information and Communications, Professor Hyungsik Kim of the College of Software Convergence, Professor Jooyoung Shin of the College of Pharmacy, Professor Jaeyeol Cho of the College of Life Sciences, Professor Sehoon Lee of the College of Medicine, Professor Youngmin Kim of the Sungkyunkwan University Convergence Institute, and Professor Honghee Won of the Samsung Institute of Convergence Medicine and Science as the '2024 SKKU-Fellowship' professors. The SKKU-Fellowship system is the highest honor that our university has awarded since 2004, and is a system that selects the best professors whose research capabilities or industry-university cooperation achievements have reached world-class standards or are highly accessible, and grants them exceptional research support and honor. The 2024 SKKU-Fellowship is based on the university management policy of “Inspiring Future, Grand Challenge” for the 23rd and 24th academic years, and selected recipients in the fields of renowned international conferences, top journals and papers, and industry-academia cooperation ecosystems by expanding the excellence and scope of each professor. The awards ceremony was held at the general faculty meeting held on Wednesday, February 19th last year. This year, the 2024 SKKU Fellowship officially recommended candidates through the Fellowship Advisory Board, which advised on the selection of candidates, and Director of Industry-Academia Cooperation Gu Ja-chun, a member of the Fellowship Advisory Board, personally announced the 10 recipients. ▲ (From the top left) Professor Kim Min-woo, Professor Seo Ah-young, Professor Son Dong-hee, Professor Choi Jae-hyeok, Professor Kim Hyung-sik, Professor Shin Joo-young, Professor Cho Jae-yeol, Professor Kim Young-min Professor Kim Min-woo of the College of Social Sciences and Professor Kim Young-min of the Sungkyunkwan University of Convergence gave their acceptance speeches on behalf of the awardees. In the future, our university will continue to discover the various excellent achievements and values of the best professors and move forward to become a first-class university that contributes to human society.
-
- 작성일 2025-03-04
- 조회수 1114