Seok-Hee Hong

Amirmohammad Pasdar, Young Choon Lee, Seok-Hee Hong

The vulnerability of smartphones to cyberattacks has been a severe concern to users arising from the integrity of installed applications (\textit{apps}). Although applications are to provide legitimate and diversified on-the-go services, harmful and dangerous ones have also uncovered the feasible way to penetrate smartphones for malicious behaviors. Thorough application analysis is key to revealing malicious intent and providing more insights into the application behavior for security risk assessments. Such in-depth analysis motivates employing deep neural networks (DNNs) for a set of features and patterns extracted from applications to facilitate detecting potentially dangerous applications independently. This paper presents an Analytic-based deep neural network, Android Malware detection (ADAM), that employs a fine-grained set of features to train feature-specific DNNs to have consensus on the application labels when their ground truth is unknown. In addition, ADAM leverages the transfer learning technique to obtain its adjustability to new applications across smartphones for recycling the pre-trained model(s) and making them more adaptable by model personalization and federated learning techniques. This adjustability is also assisted by federated learning guards, which protect ADAM against poisoning attacks through model analysis. ADAM relies on a diverse dataset containing more than 153000 applications with over 41000 extracted features for DNNs training. The ADAM's feature-specific DNNs, on average, achieved more than 98% accuracy, resulting in an outstanding performance against data manipulation attacks.

Seokhee Hong, Sunkyoung Kim, Guijin Son et al. · allen-ai, anthropic

The development of Large Language Models (LLMs) requires robust benchmarks that encompass not only academic domains but also industrial fields to effectively evaluate their applicability in real-world scenarios. In this paper, we introduce two Korean expert-level benchmarks. KMMLU-Redux, reconstructed from the existing KMMLU, consists of questions from the Korean National Technical Qualification exams, with critical errors removed to enhance reliability. KMMLU-Pro is based on Korean National Professional Licensure exams to reflect professional knowledge in Korea. Our experiments demonstrate that these benchmarks comprehensively represent industrial knowledge in Korea. We release our dataset publicly available.

Hwaran Lee, Seokhee Hong, Joonsuk Park et al.

The potential social harms that large language models pose, such as generating offensive content and reinforcing biases, are steeply rising. Existing works focus on coping with this concern while interacting with ill-intentioned users, such as those who explicitly make hate speech or elicit harmful responses. However, discussions on sensitive issues can become toxic even if the users are well-intentioned. For safer models in such scenarios, we present the Sensitive Questions and Acceptable Response (SQuARe) dataset, a large-scale Korean dataset of 49k sensitive questions with 42k acceptable and 46k non-acceptable responses. The dataset was constructed leveraging HyperCLOVA in a human-in-the-loop manner based on real news headlines. Experiments show that acceptable response generation significantly improves for HyperCLOVA and GPT-3, demonstrating the efficacy of this dataset.

Shijun Cai, Seok-Hee Hong, Jialiang Shen et al.

Understanding what graph layout human prefer and why they prefer is significant and challenging due to the highly complex visual perception and cognition system in human brain. In this paper, we present the first machine learning approach for predicting human preference for graph layouts. In general, the data sets with human preference labels are limited and insufficient for training deep networks. To address this, we train our deep learning model by employing the transfer learning method, e.g., exploiting the quality metrics, such as shape-based metrics, edge crossing and stress, which are shown to be correlated to human preference on graph layouts. Experimental results using the ground truth human preference data sets show that our model can successfully predict human preference for graph layouts. To our best knowledge, this is the first approach for predicting qualitative evaluation of graph layouts using human preference experiment data.