Sangjun Lee

Inwoo Hwang, Sangjun Lee, Yunhyeok Kwak et al.

Neural networks trained with ERM (empirical risk minimization) sometimes learn unintended decision rules, in particular when their training data is biased, i.e., when training labels are strongly correlated with undesirable features. To prevent a network from learning such features, recent methods augment training data such that examples displaying spurious correlations (i.e., bias-aligned examples) become a minority, whereas the other, bias-conflicting examples become prevalent. However, these approaches are sometimes difficult to train and scale to real-world data because they rely on generative models or disentangled representations. We propose an alternative based on mixup, a popular augmentation that creates convex combinations of training examples. Our method, coined SelecMix, applies mixup to contradicting pairs of examples, defined as showing either (i) the same label but dissimilar biased features, or (ii) different labels but similar biased features. Identifying such pairs requires comparing examples with respect to unknown biased features. For this, we utilize an auxiliary contrastive model with the popular heuristic that biased features are learned preferentially during training. Experiments on standard benchmarks demonstrate the effectiveness of the method, in particular when label noise complicates the identification of bias-conflicting examples.

Sangjun Lee

The increasing necessity for fisheye cameras in fields such as robotics and autonomous driving has led to the proposal of various fisheye camera models. While the evolution of camera models has facilitated the development of diverse systems in the field, the lack of adaptation between different fisheye camera models means that recalibration is always necessary, which is cumbersome. This paper introduces a conversion tool for various previously proposed fisheye camera models. It is user-friendly, simple, yet extremely fast and accurate, offering conversion capabilities for a broader range of models compared to existing tools. We have verified that models converted using our system perform correctly in applications such as SLAM. By utilizing our system, researchers can obtain output parameters directly from input parameters without the need for an image set and any recalibration processes, thus serving as a bridge across different fisheye camera models in various research fields. We provide our system as an open source tool available at: https://github.com/eowjd0512/fisheye-calib-adapter

Sangjun Lee, Seung-taek Woo, Jungyu Jin et al.

To enable broader deployment of Large Language Models (LLMs), it is essential to identify the best-performing model under strict memory constraints. We present AMQ, Automated Mixed-Precision Weight-Only Quantization, a framework that assigns layer-wise quantization bit-widths to optimally balance model quality and memory usage. However, the combinatorial search space, with over 10^{100} possible configurations, makes conventional black-box optimization infeasible. AMQ overcomes this challenge through four key innovations:(1) search space pruning using prior knowledge to exclude unpromising configurations, (2) quantization proxy to bypass costly format conversions during search, (3) quality predictor to minimize evaluation overhead, and (4) iterative search-and-update strategy for fast and stable convergence. By integrating these components, AMQ efficiently explores the quality-efficiency landscape, reaching the Pareto frontier and yielding LLMs that are both compact and high-performing. Our code is available at https://github.com/dlwns147/amq.

Daewoo Park, Suho Park, Inseok Hong et al.

We present AI PB, a production-scale generative agent deployed in real retail finance. Unlike reactive chatbots that answer queries passively, AI PB proactively generates grounded, compliant, and user-specific investment insights. It integrates (i) a component-based orchestration layer that deterministically routes between internal and external LLMs based on data sensitivity, (ii) a hybrid retrieval pipeline using OpenSearch and the finance-domain embedding model, and (iii) a multi-stage recommendation mechanism combining rule heuristics, sequential behavioral modeling, and contextual bandits. Operating fully on-premises under Korean financial regulations, the system employs Docker Swarm and vLLM across 24 X NVIDIA H100 GPUs. Through human QA and system metrics, we demonstrate that grounded generation with explicit routing and layered safety can deliver trustworthy AI insights in high-stakes finance.

Seungeun Lee, Il-Youp Kwak, Kihwan Lee et al.

Recent advancements in deep learning for tabular data have shown promise, but challenges remain in achieving interpretable and lightweight models. This paper introduces Table2Image, a novel framework that transforms tabular data into realistic and diverse image representations, enabling deep learning methods to achieve competitive classification performance. To address multicollinearity in tabular data, we propose a variance inflation factor (VIF) initialization, which enhances model stability and robustness by incorporating statistical feature relationships. Additionally, we present an interpretability framework that integrates insights from both the original tabular data and its transformed image representations, by leveraging Shapley additive explanations (SHAP) and methods to minimize distributional discrepancies. Experiments on benchmark datasets demonstrate the efficacy of our approach, achieving superior accuracy, area under the curve, and interpretability compared to recent leading deep learning models. Our lightweight method provides a scalable and reliable solution for tabular data classification.

Donggeun Ko, Sangjun Lee, Jinyong Park et al.

Hyper-realistic face image generation and manipulation have givenrise to numerous unethical social issues, e.g., invasion of privacy,threat of security, and malicious political maneuvering, which re-sulted in the development of recent deepfake detection methods with the rising demands of deepfake forensics. Proposed deepfake detection methods to date have shown remarkable detection performance and robustness. However, none of the suggested deepfake detection methods assessed the performance of deepfakes with the facemask during the pandemic crisis after the outbreak of theCovid-19. In this paper, we thoroughly evaluate the performance of state-of-the-art deepfake detection models on the deepfakes with the facemask. Also, we propose two approaches to enhance the masked deepfakes detection: face-patch and face-crop. The experimental evaluations on both methods are assessed through the base-line deepfake detection models on the various deepfake datasets. Our extensive experiments show that, among the two methods, face-crop performs better than the face-patch, and could be a train method for deepfake detection models to detect fake faces with facemask in real world.

Sangjun Lee, Byung-Cheol Min

This research proposes a distributed switching control to secure multi-robot systems in the presence of cyberattacks. Two major types of cyberattack are considered: deception attack and denial of service (DoS) attack, which compromise the integrity and availability of resources, respectively. First, a residual-based attack detection scheme is introduced to identify the type of attacks. Then, a switching control is designed to neutralize the effect of the identified attacks, satisfying the performance guarantees required for state consensus among robots. For the type of a deception attack, coordination-free consensus protocols are designed to tune the weights of each robot in a way that uncompromised robots gain more weight than compromised robots. For the type of a DoS attack, leader-follower protocols that reconfigure the communication topology are utilized to transform the compromised robots into sub-robots following the leaders. The performance of the proposed approach is evaluated on the Robotarium multi-robot testbed. A full demonstration with extensive cases is available at https://youtu.be/eSj0XS2pdxI.

Byung-Cheol Min, Ramviyas Parasuraman, Sangjun Lee et al.

In this paper, we present a directional antenna-based leader-follower robotic relay system capable of building end-to-end communication in complicated and dynamically changing environments. The proposed system consists of multiple networked robots - one is a mobile end node and the others are leaders or followers acting as radio relays. Every follower uses directional antennas to relay a communication radio and to estimate the location of the leader robot as a sensory device. For bearing estimation, we employ a weight centroid algorithm (WCA) and present a theoretical analysis of the use of WCA for this work. Using a robotic convoy method, we develop online, distributed control strategies that satisfy the scalability requirements of robotic network systems and enable cooperating robots to work independently. The performance of the proposed system is evaluated by conducting extensive real-world experiments that successfully build actual communication between two end nodes.

Sangjun Lee, Yongbum Cho, Byung-Cheol Min

In this paper, we propose a fault detection and isolation based attack-aware multi-sensor integration algorithm for the detection of cyberattacks in autonomous vehicle navigation systems. The proposed algorithm uses an extended Kalman filter to construct robust residuals in the presence of noise, and then uses a parametric statistical tool to identify cyberattacks. The parametric statistical tool is based on the residuals constructed by the measurement history rather than one measurement at a time in the properties of discrete-time signals and dynamic systems. This approach allows the proposed multi-sensor integration algorithm to provide quick detection and low false alarm rates for applications in dynamic systems. An example of INS/GNSS integration of autonomous navigation systems is presented to validate the proposed algorithm by using a software-in-the-loop simulation.