187037 research outputs found
Sort by
SAIF: A Comprehensive Framework for Evaluating the Risks of Generative AI in the Public Sector
No full textThe rapid adoption of generative AI in the public sector, encompassing diverse applications ranging from automated public assistance to welfare services and immigration processes, highlights its transformative potential while underscoring the pressing need for thorough risk assessments. Despite its growing presence, evaluations of risks associated with AI-driven systems in the public sector remain insufficiently explored. Building upon an established taxonomy of AI risks derived from diverse government policies and corporate guidelines, we investigate the critical risks posed by generative AI in the public sector while extending the scope to account for its multimodal capabilities. In addition, we propose a Systematic dAta generatIon Framework for evaluating the risks of generative AI (SAIF). SAIF involves four key stages: breaking down risks, designing scenarios, applying jailbreak methods, and exploring prompt types. It ensures the systematic and consistent generation of prompt data, facilitating a comprehensive evaluation while providing a solid foundation for mitigating the risks. Furthermore, SAIF is designed to accommodate emerging jailbreak methods and evolving prompt types, thereby enabling effective responses to unforeseen risk scenarios. We believe that this study can play a crucial role in fostering the safe and responsible integration of generative AI into the public sector
UNSUPERVISED DOMAIN ADAPTATION SYSTEM AND METHOD
No full text본 문서에 개시되는 일 실시예에 따른 비지도 도메인 적응 시스템은, 라벨링 정보를 포함하는 소스 도메인 및 라벨링 정보를 포함하지 않는 타겟 도메인을 입력 받아 소스 벡터 및 타겟 벡터를 추출하되, 상기 소스 벡터와 상기 타겟 벡터가 특성을 서로 공유하도록 상기 소스 벡터 및 상기 타겟 벡터를 추출하는 과정을 학습하고, 상기 추출하는 과정에서 획득되는 상기 타겟 도메인의 타겟 특징을 저장하는 제1학습부; 및 상기 타겟 벡터가 상기 소스 도메인 및 상기 타겟 도메인 중 어느 도메인으로부터 추출된 벡터인지를 판별하는 과정을 학습하고, 판별 결과를 제공하는 제2 학습부를 포함할 수 있다
AVATAR-BASED TELEPRESENCE SYSTEM FOR GENERATING DEICTIC MOTION AND THE OPERATING METHOD THEREOF
No full text본 발명은 사용자의 지시동작을 파악하여 해당 의미를 다른 공간에 위치한 증강현실 텔레프레즌스 아바타가 재현할 수 있도록 아바타의 지시동작을 생성하는 텔레프레즌스 시스템 및 그 동작 방법에 관한 것으로서, 제1 사용자의 지시동작의 완료상태(CS)를 탐지하는 지시동작 탐지모듈, 상기 제1 사용자의 행동으로 신경망 기반의 프레임워크를 학습시켜 제1 아바타의 지시동작을 산출하는 지시동작 계산모듈 및 상기 제1 아바타의 지시동작을 상기 제1 아바타가 수행하도록 하기 위해 다음 프레임의 애니메이션으로 생성하는 동작 생성 모듈을 포함한다
One Look is Enough: Seamless Patchwise Refinement for Zero-Shot Monocular Depth Estimation on High-Resolution Images
No full text.Zero-shot depth estimation (DE) models exhibit strong generalization performance as they are trained on large-scale datasets. However, existing models struggle with high-resolution images due to the discrepancy in image resolutions of training (with smaller resolutions) and inference (for high resolutions). Processing them at full resolution leads to decreased estimation accuracy on depth with tremendous memory consumption, while downsampling to the training resolution results in blurred edges in the estimated depth images. Prevailing high-resolution depth estimation methods adopt a patch-based approach, which introduces depth discontinuity issues when reassembling the estimated depth patches and results in test-time inefficiency. Additionally, to obtain fine-grained depth details, these methods rely on synthetic datasets due to the real-world sparse ground truth depth, leading to poor generalizability. To tackle these limitations, we propose Patch Refine Once (PRO), an efficient and generalizable tile-based framework. Our PRO consists of two key components: (i) Grouped Patch Consistency Training that enhances test-time efficiency while mitigating the depth discontinuity problem by jointly processing four overlapping patches and enforcing a consistency loss on their overlapping regions within a single backpropagation step, and (ii) Bias-Free Masking that prevents the DE models from overfitting to dataset-specific biases, enabling better generalization to real-world datasets even after training on synthetic data. Zero-shot evaluation on Booster, ETH3D, Middlebury 2014, and NuScenes demonstrates into which our PRO can be well harmonized, making their DE capabilities still effective for the grid input of high-resolution images with little depth distinuities at the grid boundaries. Our PRO runs fast at inference time
기하학적 오브젝트 생성을 위한 제너레이티브 플로우 네트워크
No full text학위논문(석사) - 한국과학기술원 : 김재철AI대학원, 2025.2,[iv, 22 p. :]We employ GFlowNets for 3D geometric object generation. In order to establish empirical knowledge about the functionality and performance of the GFlowNets framework in geometric object generation tasks, which lie in continuous domains, we create a 3D point cloud environment and train GFlowNets on a benchmark 3D object dataset. The experiments show that the continuous GFlowNets model achieves similar performance compared to known methods, and discovers all modes of the target data distribution.한국과학기술원 :김재철AI대학원
도시 범죄의 해결책 또는 촉매제로서의 마이크로모빌리티:시카고 사례를 중심으로
No full text학위논문(석사) - 한국과학기술원 : 기술경영학부, 2025.2,[iii, 40 p. :]Micromobility, referring to lightweight vehicles such as e-scooters, is gaining attention as a promising urban transportation solution due to its benefits in improving connectivity with public transportation and reducing traffic congestion. However, there is ongoing debate regarding its impact on crime. This study aims to examine the effect of micromobility on urban crime, drawing on Routine Activity Theory, which explains crime reduction, and the Perverse Effects Theory, which supports crime increase. Using a difference-in-differences approach, the study analyzes crime data from Chicago, focusing on property crimes and violent crimes, with the e-scooter pilot program serving as a reference point for the introduction of micromobility. The results show a significant increase in property crimes such as theft, with this increase being particularly pronounced during nighttime. In conclusion, the study supports the Perverse Effects Theory.한국과학기술원 :기술경영학부
인도네시아의 새로운 교육 정책 영향 탐구: ‘메르데카 벨라자르’ 사례
No full text학위논문(석사) - 한국과학기술원 : 기술경영학부, 2025.2,[iii, 28 p. :]The Merdeka Belajar policy, introduced in 2020 by Indonesia’s Ministry of Education, seeks to reform the education system through decentralization, digital integration and targeted financial support for underserved students. This study assesses the policy's effectiveness in improving secondary school enrollment and advancing human development outcomes (HDI), focusing on the roles of education expenditure and digital accessibility. Employing data from 1999 to 2023 and Ordinary Least Squares regression models, the analysis confirms that increased education spending and expanded digital infrastructure significantly boost secondary school enrollment. Moreover, secondary school enrollment positively influences HDI. These findings underscore the importance of long-term investments in education and technology to reduce disparities and support inclusive development.한국과학기술원 :기술경영학부
메탄 산화 반응에서 저온 활성 및 높은 내구성을 위한 팔라듐-세리아-알루미나 촉매 연구
No full text학위논문(석사) - 한국과학기술원 : 생명화학공학과, 2025.2,[iv, 25 p. :]Methane, a greenhouse gas with 28 times the global warming potential of CO₂, significantly contributes to climate change and air pollution. Despite the adoption of liquefied natural gas (LNG) as a cleaner alternative in the shipping industry, methane slip undermines its environmental benefits. Pd-based catalysts have shown potential for complete methane oxidationhowever, their limited low-temperature activity and susceptibility to moisture-induced deactivation remain significant challenges. This study addresses these issues by optimizing temperature control and catalyst design to enhance catalyst performance under wet conditions. Temperature regulation was improved by optimizing diluents and cooling rates, mitigating hotspot formation and enabling accurate evaluations of catalytic activity. Two Pd/CeO₂-Al₂O₃ synthesis methods were compared, and methane oxidation tests were conducted under dry and wet conditions to assess reactivity and resistance to moisture-induced deactivation. Cyclic reaction tests revealed differences in activity saturation behaviors between dry and wet environments under varying space velocities. Structural and electronic characterizations of catalysts before and after saturation provided insights into the mechanisms driving catalytic performance. These findings contribute to the development of robust and efficient Pd-based catalysts for effective methane emission control under realistic operating conditions.한국과학기술원 :생명화학공학과
음이온 교환막 수전해 장치에서 부하 변동과 역전류에 의한 백금 합금 촉매 열화 연구
No full text학위논문(석사) - 한국과학기술원 : 생명화학공학과, 2025.2,[iv, 38 p. :]Hydrogen energy is gaining attention as a clean and sustainable alternative to fossil fuels, and anion exchange membrane water electrolyzer (AEMWE) have emerged as a promising next-generation hydrogen production technology due to their ability to utilize non-precious materials and operate at low cost. However, platinum (Pt) based catalysts widely used in the cathodes of AEMWE face challenges such as cathode oxidation and catalyst degradation during frequent start-up and shut-down cycles, leading to reduced performance and durability. This study investigates the degradation mechanisms of platinum alloy catalysts and evaluates their stability and performance under various operating conditions to explore their potential as replacements for commercial catalysts. The findings aim to provide critical insights for ensuring long-term stability and improving catalyst performance, contributing to the development of sustainable hydrogen production technologies.한국과학기술원 :생명화학공학과
링커 유형에 따른 이량체 억셉터 구조 개질을 통한 고성능 유기 태양전지 구현
No full text학위논문(석사) - 한국과학기술원 : 생명화학공학과, 2025.2,[iv, 46 p. :]The dimerization of small molecule acceptors (DSMAs) has gained attention as an effective strategy to enhance the long-term stability and power conversion efficiency (PCE) of organic solar cells (OSCs). However, most studies focus on end-linked configurations, leaving other designs underexplored. This study introduces two head-type dimerized SMAs, DYF-V and DYF-E, with distinct linker types (vinylene and ethynyl, respectively). DYF-E, featuring an ethynyl linker, exhibits a twisted backbone and reduced aggregation behavior, leading to superior blend morphologies with polymer donors, including high crystallinity and well-mixed domains. As a result, DYF-E-based OSCs achieve a high PCE of 17.02%, outperforming DYF-V (PCE = 9.98%), with ternary OSCs incorporating DYF-E reaching an impressive 18.53%. These findings highlight the critical impact of linker selection on dimer properties and OSC performance. Furthermore, this study not only expands the structural diversity of dimerized SMAs but also provides valuable insights into the design principles for achieving optimal blend morphologies and high PCEs in OSCs.한국과학기술원 :생명화학공학과