GIST Scholar
Not a member yet
30271 research outputs found
Sort by
Data-free Universal Adversarial Perturbation with Pseudo-semantic Prior
No full textData-free Universal Adversarial Perturbation (UAP) is an image-agnostic adversarial attack that deceives deep neural networks using a single perturbation generated solely from random noise without relying on data priors. However, traditional data-free UAP methods often suffer from limited transferability due to the absence of semantic content in random noise. To address this issue, we propose a novel data-free universal attack method that recursively extracts pseudo-semantic priors directly from the UAPs during training to enrich the semantic content within the data-free UAP framework. Our approach effectively leverages latent semantic information within UAPs via region sampling, enabling successful input transformations - typically ineffective in traditional data-free UAP methods due to the lack of semantic cues - and significantly enhancing black-box transferability. Furthermore, we introduce a sample reweighting technique to mitigate potential imbalances from random sampling and transformations, emphasizing hard examples less affected by the UAPs. Comprehensive experiments on ImageNet show that our method achieves state-of-the-art performance in average fooling rate by a substantial margin, notably improves attack transferability across various CNN architectures compared to existing data-free UAP methods, and even surpasses data-dependent UAP methods. Code is available at: https://github.com/ChnanChan/PSP-UAP. © 2025 IEEE
Robust Object Detection under Adverse Conditions via Joint Semantic Learning and Knowledge Distillation
No full text인식 부족, 가혹한 환경 조건에 대한 취약성, 도메인 변화에 대한 일반화 한계 등의 문제를 안고 있다. 이를 해결하기 위해, 본 연구에서는 대규모 다중 시점 해양 데이터셋, 강건한 해양 객체 탐지를 위한 공동 학습 프레임워크, 일반화 성능을 향상시키는 교차 도메인지식증류기법을제안한다. 첫째,우리는 Multi-View Maritime Vision (MV2)데이터셋을소개한다. MV2는선박 주변 6개시점에서촬영된 159,386개의가시광이미지로구성되며, 360도전방위환경인 식을지원한다.본데이터셋은객체바운딩박스,의미론적,인스턴스,파놉틱분할라벨을 포함하며, 다양한 환경 조건을 반영하여 해양 상황 인식 연구에 적합하도록 설계되었다. 또한, 최신 객체 탐지 및 파놉틱 분할 모델을 MV2에서 벤치마킹하여 해양 자율성 연구 발전에기여할수있음을실증하였다. 둘째, Joint Semantic Learning (JSL)을제안하여해양환경에서의강건한객체탐지를 가능하게 하는 프레임워크를 개발하였다. JSL은 해양 장면 분할과 객체 탐지를 결합하 여 탐지 네트워크의 전반적인 성능을 향상시킨다. 학습 시 분할 모듈을 추가하여 장면의 맥락을 학습하도록 유도하지만, 추론 시에는 분할 모듈을 제거하여 추가적인 연산 부담 없이 동작한다. 실험 결과, JSL을 적용한 모델은 SMD 및 SeaShips 데이터셋에서 높은 – iii – 성능향상을보였으며,특히안개,조명변화,모션블러등가혹한조건에서도뛰어난탐 지성능을유지하는것으로나타났다.또한, Cityscapes데이터셋을활용한실험에서도시 환경에서도효과적으로일반화됨을확인하였다. 셋째,우리는 Cross-Domain Feature Knowledge Distillation (CD-FKD)을제안하여도메 인변화에대한객체탐지모델의일반화성능을향상시킨다. CD-FKD는전역및인스턴 스단위의특징증류를활용하며,학생네트워크는손상된및축소된데이터를학습하고, 교사 네트워크는 원본 도메인 데이터를 학습하도록 설계되었다. 이 과정에서 학생 네트 워크는객체중심특징을보다효과적으로추출할수있으며,도메인변화에강인한탐지 성능을 갖출 수 있다. 다양한 실험을 통해 CD-FKD가 최첨단 방법론을 뛰어넘는 성능을 보이며,자율주행및감시시스템과같은도메인일반화가필수적인실제응용분야에서 효과적으로활용될수있음을검증하였다. 본연구에서제안하는기법들은다중시점인식,가혹한환경에서도강건한탐지성능, 도메인 변화에 대한 일반화 능력 향상이라는 핵심 문제를 해결한다. 이러한 연구 성과는 해양,교통,감시분야의자율시스템에서신뢰할수있는객체탐지를구현하는데중요한 기여를할것으로기대된다. |Reliable object detection is critical for autonomous systems operating in dynamic en- vironments, yet existing models face challenges in comprehensive multi-view perception, robustness to adverse conditions, and generalization across domain shifts. To address these limitations, we propose three key advancements: a large-scale multi-view maritime dataset, a joint learning framework for robust maritime object detection, and a cross-domain feature knowledge distillation method for improved generalization. First, we introduce the Multi-View Maritime Vision (MV2) dataset, comprising 159,386 visible-light images captured from six viewpoints around a vessel, providing a complete 360-degree perception. MV2 includes bounding boxes, semantic, instance, and panoptic segmentation labels, covering diverse environmental conditions to support maritime situational awareness. Benchmarks on state-of-the-art models validate its utility for advancing maritime autonomy research. Second, we present Joint Semantic Learning (JSL), a framework that enhances object detection in challenging maritime environments by integrating ocean scene segmentation into the detection network. JSL enables the model to extract both spatial and semantic information, improving robustness to fog, lighting variations, and motion blur. Extensive experiments on SMD and SeaShips datasets demonstrate that JSL significantly enhances detection performance, particularly under adverse conditions. Furthermore, experiments on Cityscapes confirm its ability to generalize to urban environments. Third, we propose Cross-Domain Feature Knowledge Distillation (CD-FKD) to enhance model generalization in single-domain training settings. CD-FKD leverages global and instance-wise feature distillation, training the student network with downscaled and cor- rupted data while maintaining the teacher network on the original domain. This approach enables the student to learn robust object-centric features, improving performance on unseen target domains. Experiments demonstrate that CD-FKD outperforms state-of-the-art methods in both source and target domain evaluations, making it particularly beneficial for applications in autonomous driving and surveillance. Together, these contributions provide a foundation for enhanced object detection in real- world scenarios, addressing the challenges of multi-view perception, robustness to adverse conditions, and domain generalization. Our findings offer promising directions for improving autonomy in maritime, transportation, and surveillance applications.DoctorAbstract (English) i
Abstract (Korean) iii
List of Contents v
List of Tables vii
List of Figures ix
1 Introduction 1
2 A Large Scale 360-degree Multi-View Maritime Vision Dataset for Object Detection and Segmentation 6
2.1 Related Work 8
2.2 MV2 DATASET 12
2.2.1 Data Collection 12
2.2.2 Dataset Structure and Synchronization 13
2.2.3 Data Analysis 14
2.2.4 Data Annotations for Object Detection 15
2.2.5 Data Annotations for Segmentation 17
2.3 Experiments 19
2.3.1 Benchmark Performance for Object Detection 19
2.3.2 Benchmark Performance for Panoptic Segmentation 22
2.3.3 Effectiveness Verification of MV2 22
3 Robust Maritime Object Detection under Adverse Conditions via Joint Semantic Learning without Extra Computational Overhead 25
3.1 Related Work 28
3.1.1 Object Detection 28
3.1.2 Semantic Segmentation 29
3.1.3 Joint Learning 30
– v –
3.2 Method 30
3.2.1 Feature extractor 31
3.2.2 Ocean Scene Semantic Segmentation of JSL 31
3.2.3 Maritime Object Detection of JSL 32
3.2.4 Joint Loss Function 34
3.3 Experiment 35
3.3.1 Dataset and Generated Semantic Label 35
3.3.2 Maritime Object Detection Experiments 37
3.3.3 Robustness Test of JSL 40
3.3.4 Generalization Verification of JSL using Cityscapes 41
4 Cross-Domain Feature Knowledge Distillation for Robust Single-Domain Generalization in Object Detection 42
4.1 Related Work 46
4.1.1 Single Domain Generalization 46
4.1.2 Knowledge Distillation for Object Detection 47
4.2 Method 48
4.2.1 Cross Domain Distillation Framework 48
4.2.2 Diversified Source Domain Data 51
4.2.3 Global Feature Distillation 53
4.2.4 Instance-Wise Feature Distillation 54
4.3 Experiments 56
4.3.1 Experimental Setup 56
4.3.2 Comparison with State-of-the-arts 57
4.3.3 Ablation Study 61
4.3.4 Visualization Analysis 62
Summary 64
References 66
Acknowledgements 88
– vi
Voltage contrast inspection of an electrical defect on a semiconductor wafer with X-ray photoelectron spectroscopy and an electron beam
No full textHighly efficient semiconductor metrology and inspection techniques have played important roles for the early detection of defects in the semiconductor-fabrication industry. We here report a new voltage contrast inspection technique based on X-ray photoelectron spectroscopy (XPS) and an electron beam (XPS/EB-VCI). This technique detects an electrical defect (in particular, an electrically open defect) on a semiconductor wafer while simultaneously analyzing the chemical state of the wafer. In XPS/EB-VCI, the existence of an electrical defect is identified in a wide area on the sub-millimeter scale, and the location of the defect can be specified with micrometerlevel precision. Furthermore, with the application of Auger electron spectroscopy, the identified electrical defect area can be further mapped on the nanometer scale. We believe that XPS/EB-VCI could become a viable assessment tool for the semiconductor inspection process, along with a current inspection technique based on scanning electron microscopy.FALSEsciescopu
LMI-based Neural Network Observer for State and Nonlinearity Estimation
No full textThis paper proposes a design method for linear matrix inequality (LMI)-based neural network (NN) observer gain in discrete-time domain. The proposed scheme employs an NN with a single hidden layer to approximate the lumped nonlinear term which includes uncertainties. A Lyapunov function is constructed to guarantee the stability of both the linear observer and the NN updates. The observer gain is determined by solving the LMI conditions, and the design is simplified by minimizing the number of tuning parameters, using a common gain structure for all vertices. Furthermore, designing an H∞ observer can reduce the effect of NN approximation error and the measurement noise.The key advantages of the proposed method lie in its optimal LMI-based observer gain design, minimal tuning parameter requirement, and the capability to estimate both the system states and the lumped nonlinear term simultaneously. Simulation results indicate that the proposed method successfully tracks the actual states and the lumped nonlinear term and reduce the effects of NN approximation error and the measurement noise with comparison of the root mean square error (RMSE) values. © 2025 IEEE
Observational analysis of long-term streamflow response to flash drought in the Mississippi River Basin
No full textFlash droughts, characterized by rapid onset and severe intensity, pose significant challenges to water resource management. This study investigates the relationship between meteorological conditions and streamflow dynamics during flash drought events within the Mississippi River Basin (MRB) over the period 1980–2022. The MRB has been a hot spot for flash drought events in recent years with significant ecological, agricultural, and economic damages. Existing literature has predominantly focused on studying the meteorological conditions preceding and during flash drought events. However, the media attention on low flows during flash drought events in 2022 and 2023 in the MRB has highlighted the adverse impact of flash droughts on streamflow levels, emphasizing the need to study the connection between flash drought conditions and streamflow dynamics. This study employs the Standardized Antecedent Precipitation Evapotranspiration Index (SAPEI), a daily drought index, to identify flash droughts at the catchment scale and assess the response of streamflow to the identified events across the MRB. Over 1000 flash drought events were identified at the catchment scale. The analysis of these events reveals spatial variations in the frequency, intensity, and duration of flash droughts. The eastern MRB experiences frequent but shorter flash droughts, while the northwest experiences lengthier but less frequent events. The southern region grapples with the most severe drought conditions and is heavily influenced by upstream water management practices. The analysis between SAPEI and streamflow anomalies showed a strong positive correlation in 221 of the 258 catchments with an average ρ of 0.64. The SAPEI-based method for identifying flash droughts and the subsequent comparison with streamflow anomalies is highly transferable, providing a valuable framework for assessing the impacts of flash drought on streamflow and informing water management strategies in other regions. This research significantly contributes to advancing our understanding of flash droughts, offering critical insights for informed water resource management, and bolstering resilience against the impacts of flash droughts. © 2025TRUEsciescopu
An evolutionarily conserved cation channel tunes the sensitivity of gustatory neurons to ephaptic inhibition in Drosophila
No full textIn ephaptic coupling, physically adjacent neurons influence one another’s activity via the electric fields they generate. To date, the molecular mechanisms that mediate and modulate ephaptic coupling’s effects remain poorly understood. Here, we show that the hyperpolarization-activated cyclic nucleotide–gated (HCN) channel lateralizes the potentially mutual ephaptic inhibition between Drosophila gustatory receptor neurons (GRNs). While sweet-sensing GRNs (sGRNs) engage in ephaptic suppression of the adjacent bitter-sensing GRNs (bGRNs), HCN expression in sGRNs enables them to resist ephaptic suppression from the bGRNs. This one-sided ephaptic inhibition confers sweetness dominance, facilitating ingestion of bitter-laced sweets. The role of fly HCN in this process can be replaced by human HCN2. Furthermore, unlike the mechanism in olfaction, gustatory ephaptic inhibition is independent of sensillum potential changes, suggesting that the compartmentalized arrangement of neighboring GRNs is dispensable for gustatory ephaptic inhibition. These findings indicate a role for the gating of ephaptic coding to ensure the intake of the essential nutrient despite bitter contaminants present in the feeding niche of Drosophila, and propose that studies in Drosophila gustation could reveal ephaptic principles conserved across diverse animals. Copyright © 2025 the Author(s).TRUEsciescopu
비백금 음이온 교환막 수전해를 위한 철 도핑 활용 니 켈-몰리브데넘 합금 촉매의 활성점 제어
No full textDeveloping advanced catalysts for the hydrogen evolution reaction in alkaline media remains a significant hurdle for the broader adoption of anion exchange membrane water electrolyzers. This study presents the synthesis of an Fe-doped Ni4Mo/MoOx catalyst through a synergistic approach involving hydrothermal and thermal reduction processes. Material characterization and electron structure analysis underscored the pivotal role of Fe, and its addition notably promoted the formation of the Ni4Mo phase, which was identified as active site of the catalyst. Further insights from density functional theory calculations revealed that Fe doping enhanced the desorption energy necessary for the H2 recombination step within the Ni4Mo phase, bolstering both the intrinsic and extrinsic activities. This enhancement propelled the NiMoFe catalyst to outperform conventional NiMo catalysts in both aqueous model system and membrane electrode assembly experiment. Notably, an anion exchange membrane water electrolysis configured with the NiMoFe catalyst at the cathode demonstrated remarkable performance, reaching a cell voltage of 1.63 V at a current density of 1 A cm-2, and preserving over 98% of its initial voltage at 0.5 A cm-2, demonstrating its potential for commercial applications.Master1. Introduction 1
1.1 Principal of Hydrogen evolution reaction (HER)
1.2 The advantages and challenges of Ni4Mo-based catalyst
1.3 The role of doping in enhancing electrochemical Catalyst
performance
1.4 Objectives of this research
2. Experimental and Method 5
2.1 Materials preparation
2.2 Electrochemical characterization
2.3 Structural characterization
2.4 AEMWE operation
2.5 Computational details
3. Result and Discussion. 13
4. Conclusion 31
5. Reference. 32
6. Acknowledgement 33
7. Curriculum Vitae 4
Studies on antimicrobial peptoids: effect of metal binding motif conjugation and polymer grafting Dasom Song
No full textAntimicrobial peptoids have been developed as promising alternatives to traditional antimicrobial peptides (AMPs) in response to urgent challenge of antimicrobial resistance (AMR). With enhanced stability and resistance to enzymatic degradation, peptoids offer significant advantages for therapeutic applications. Building on these benefits, numerous studies have focused on improving the activity of antimicrobial peptoids through various modifications and applications. Chapter 1 provides an overview of these advancements and highlights ongoing efforts to enhance their therapeutic potential. This chapter establishes a foundation for subsequent research, including the present study, which aims to further expand the functionality and effectiveness of antimicrobial peptoids in combating bacterial infections. Chapter 2 investigates the incorporation of catalytic metal-binding motifs—the amino terminal Cu(II) and Ni(II) binding (ATCUN) motif—into cationic amphipathic antimicrobial peptoids to enhance their activity. The ATCUN motif, upon binding to Cu(II), facilitates the generation of hydroxyl radicals through a Fenton-like reaction. These hydroxyl radicals, potent reactive oxygen species (ROS), play a crucial role in bacterial cell death. We synthesized a library of ATCUN-peptoid conjugates and evaluated their antimicrobial activity, cytotoxicity, and bacterial killing mechanisms. Additionally, we confirmed that these peptoid derivatives exhibit both antimicrobial and anti-inflammatory effects with low toxicity in a sepsis-induced mouse model, underscoring their potential as therapeutic agents. Chapter 3 focuses on the design and synthesis of antimicrobial peptoids conjugated with ZnDPA (zinc(II)-dipicolylamine) and its derivative, Zn2BPMP (BPMP; 2,6-bis[(bis(2-pyridylmethyl)amino)-methyl]-4- methylphenolate), to enhance selectivity for bacterial membranes. This chapter begins with an overview of bacterial cell membrane composition, emphasizing the anionic nature that distinguishes bacterial membranes from mammalian ones. The incorporation of ZnDPA complexes into peptoids is inteded to exploit this unique feature, enhancing antimicrobial activity while minimizing hemolytic activity against eukaryotic cells. The chapter details the design and synthesis of ZnDPA and Zn2BPMP conjugated antimicrobial peptoids to target the anionic characteristics of bacterial membranes, aiming to enhance selectivity and activity. The synthesized peptoid conjugates were evaluated through various antimicrobial assays against Gram-positive and Gram-negative bacteria, including multidrug-resistant strains. Additional studies were conducted to assess membrane interactions, hemolytic activity, and stability under physiological conditions. The findings highlight the effectiveness of ZnDPA and Zn2BPMP conjugation in improving the antimicrobial potential of peptoids, supporting their development as targeted agents against drug-resistant bacterial infections. Chapter 4 presents the synthesis and evaluation of antimicrobial peptoid-polymer conjugates (poly-PCs) designed to leverage multivalency for enhanced antimicrobial activity. Utilizing PET-RAFT polymerization, a series of antimicrobial poly-PCs with varied polymer chain lengths and peptoid content were synthesized to investigate how these structural variations influence biological activity. The antimicrobial poly-PCs were characterized through gel permeation chromatography (GPC) and NMR, followed by antimicrobial assays against E. coli and S. aureus. Hemolytic assays were also performed to assess toxicity towards mammalian cells. The findings demonstrate that shorter polymer backbones with moderate peptoid content generally show higher antimicrobial activity while balancing biocompatibility. This research underscores the potential of antimicrobial poly-PCs as customizable platforms for developing advanced antimicrobial materials with tunable properties, contributing to the broader field of peptoid conjugates for combating drug-resistant bacteria.DoctorAbstract ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ i
List of contents ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ iii
List of tables ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ vi
List of figures ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ vi
List of schemes ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․vii
Chapter 1. Introduction ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 1
1.1. Antimicrobial resistance (AMR)․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 1
1.2. Antimicrobial peptides (AMPs)․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 2
1.3. Peptoids․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ ․․․․․․․․ 4
1.4. Antimicrobial peptoids․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 6
1.2.1. Strategies to improve selectivity and potency․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 6
1.5. References․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 9
Chapter 2. The amino terminal Cu(II) and Ni(II) binding(ATCUN) motif conjugated antimicrobial peptoids․․․․ 14
2.1. Introduction․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 14
2.1.1. Hydroxyl radicals for antimicrobial activity․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 14
2.1.2. Amino terminal Cu(II) and Ni(II) binding(ATCUN) motif․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 15
2.2. Research objectives․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ ․․․ 17
2.3. Results and discussion․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 18
2.3.1. Design and synthesis of ATCUN-peptoids․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 18
2.3.2. In vitro biological evaluation ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 27
2.3.3. Circular dichroism study․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 32
2.3.4. Serum stability test of 22 and 22-Cu․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 35
2.3.5. Antimicrobial mechanism studies․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 36
2.3.6. Antimicrobial activities against streptococcus pneumonia․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 43
2.3.7. Anti-inflammatory activity․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 44
2.3.8. In vivo studies․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 48
2.3.9. Measurement of intracellular Cu(II) concentration․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 53
2.3.10. Competition for Cu(II) ions between human serum albumin (HSA) and peptoid22․ 54
2.4. Experimental․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 56
2.5. Conclusion․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ ․․․․ 68
2.6. References․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ ․․․․ 69
Chapter 3. Zinc(II)-dipicolylamine (ZnDPA) containing antimicrobial peptoids ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 76
3.1. Introduction․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 76
3.1.1. Bacterial cell membrane composition ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 76
3.1.2. Zinc(II)-dipicolylamine (ZnDPA)․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 77
3.2. Research objectives․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ ․․․ 78
3.3. Results and discussion․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 79
3.3.1. DPA/BPMP-peptoid design and synthesis․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 79
3.3.2. Antimicrobial activities․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 82
3.3.3. Hemolytic activity․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 83
3.3.4. In vitro cytotoxicity ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 85
3.3.5. Antimicrobial activity against MDR strains․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 87
3.3.6. Salt and serum stability test․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 88
3.3.7. LPS and LTA binding․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 89
3.3.8. Bacterial membrane disruption․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 90
3.4. Experimental․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 95
3.5. Conclusion․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ ․․․․ 105
3.6. References․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 106
Chapter 4. Antimicrobial peptoid-polymer conjugates (Poly-PCs) ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 110
4.1. Introduction․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 110
4.1.1. Multivalent effect ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 110
4.1.2. Antimicrobial peptide-polymer conjugates․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 111
4.1.3. Photoinduced electron/energy transfer reversible addition–fragmentation chain transfer
(PET-RAFT) polymerization․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 112
4.2. Research objectives․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 114
4.3. Results and discussion․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 115
4.3.1. Design and synthesis of antimicrobial poly-PCs using PET-RAFT polymerization․․․․․․․․․․․․․ 115
4.3.2. Antimicrobial poly-PCs with varied peptoid content ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 117
4.3.3. Antimicrobial and hemolytic activity of antimicrobial poly-PCs with varying peptoid
content․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 120
4.3.4. Synthesis of antimicrobial poly-PCs with fixed DP and varied monomer ratio․․․․․․․․ 122
4.3.5. Biological activities of antimicrobial poly-PCs with fixed DP and varied peptoid content․․․․․․ 123
4.3.6. Bacterial inner membrane disruption․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 126
4.4. Experimental․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 130
4.5. Conclusion․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 135
4.6. References․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 136
Acknowledgements․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 138
Curriculum Vitae․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 13
Joint Torque-based Sarcopenia Assessment in Activities of Daily Living Using a Data-driven Simulation Framework
No full textSarcopenia is a progressive muscle disease affecting a significant portion of the elderly population, yet conventional diagnostic methods primarily rely on indirect or subjective assessments. It is associated with increased risks of falls, disability, hospitalization, and mortality, making early and accurate diagnosis critical for preserving mobility and independence for older adults. While joint torque analysis offers a direct evaluation of muscle strength and function, traditional musculoskeletal simulations are limited by an expensive, laboratory-based, and expert-reliant methodology. Therefore, we introduce Motion-AI Integrated Surveillance for the Elderly (MAISE), a novel framework that estimates joint torque during daily activities for direct evaluation of muscle strength and function. First, a reinforcement learning method for scaling automated musculoskeletal models was introduced. This approach significantly improved model personalization and effectively reduced modeling error, surpassing existing optimization tools. Second, a method for estimating ground reaction force using only motion data was developed, incorporating a novel Center of Pressure (CoP) Limiter that embeds spatial information. CoP Limiter enables the accurate prediction of ground reaction forces without the need for expensive force plates. Third, the applicability of MAISE was demonstrated in a study of twenty-eight elderly participants, including Sarcopenia patients. Strong correlations were identified between joint torque metrics (peak torque, rate of torque development, and power range) and conventional Sarcopenia indicators. Additionally, significant differences in torque metrics were observed between Sarcopenia and healthy groups. These findings demonstrate dynamic joint torque analysis as a sensitive and objective method for diagnosing and monitoring Sarcopenia. Therefore, the proposed data-driven simulation pipeline advances the accuracy and accessibility of daily-life joint torque estimation, demonstrating its feasibility as a practical digital biomarker.|근감소증은 노인 인구의 대부분에 영향을 미치는 진행성 근육 질환이지만, 기존의 진단 방식은 주로 간접적이거나 주관적인 평가에 의존한다. 이는 낙상, 장애, 입원 및 사망 위험 증가와 연관되어 노인의 이동성과 자립성 유지를 위해 정확한 조기 진단이 중요하다. 관절 토크 분석은 근력과 기능을 직접적으로 평가할 수 있지만, 이를 추정하기 위한 전통적인 근골격 시뮬레이션은 고비용의 실험실 기반이며 전문가 의존성이라는 한계를 지닌다. 이에, 본 연구에서는 일상 활동 중 관절 토크를 추정하여 근력과 기능을 직접 평가하는 새로운 프레임워크인 노인을 위한 동작-인공지능 통합 감시 시스템(MAISE)을 제안한다. 첫째, 자동화된 근골격 모델의 스케일링을 위한 강화 학습 방법을 도입했다. 이 접근법은 기존 최적화 도구를 능가하며 모델 개인화를 위한 모델링 오류를 효과적으로 감소시켰다. 둘째, 동작 데이터만을 사용하여 지면반력을 추정하는 방법을 개발했다. 이는 공간 정보를 포함하는 새로운 압력 중심 제한기를 통합하여 고가의 힘판 없이도 정확한 지면반력 예측이 가능함을 보였다. 셋째, 프레임워크의 적용 가능성을 입증하기 위해 근감소증 환자를 포함한 28명의 노인 참여자를 대상으로 연구를 수행했다. 관절 토크 지표(최대 토크, 토크 발생 속도, 파워 범위)와 기존 근감소증 지표 간에 높은 상관관계가 확인되었고, 근감소증 그룹과 건강 그룹 간 토크 지표에서 유의미한 차이가 관찰되었다. 이러한 결과는 동적 관절 토크 분석이 근감소증을 진단하고 모니터링하는 민감하고 객관적인 방법임을 보여준다. 결론적으로, 본 논문에서 제안한 데이터 기반 시뮬레이션 파이프라인은 일상생활에서의 관절 토크 추정의 정확성과 접근성을 개선하여, 관절 토크를 일상적 디지털 바이오마커로 확장할 수 있는 실질적 가능성을 제시한다.MasterI Introduction 1
II Musculoskeletal Simulation with OpenSim 3
2.1 Background and Related Works 3
2.2 Musculoskeletal Model 4
2.3 OpenSim Pipeline 6
2.3.1 Model Scaling 6
2.3.2 Inverse Kinematics 8
2.3.3 Inverse Dynamics 9
2.3.4 Residual Reduction Algorithm 10
2.4 Application: Computing Ground Reaction Forces 11
III Reinforcement-Learning-Driven Automatic Scaling 19
3.1 Background and Related Works 19
3.2 Methods 21
3.2.1 Problem Formulation 21
3.2.2 Environment Description 23
3.2.3 Agent 27
3.2.4 Training and Evaluation Protocol 30
3.3 Results 31
3.3.1 Pre-training on the OpenCap cohort 31
3.3.2 Subject-specific fine-tuning on the elderly cohort 32
3.4 Discussion 34
3.4.1 Main Findings 34
3.4.2 Limitations and Future Directions 36
IV CoP-Constrained Ground Reaction Force Estimation 38
4.1 Background and Related Works 38
4.2 Methods 39
4.2.1 CoP-Constrained Learning 39
4.2.2 Model Architecture 42
4.2.3 Training 43
4.3 Results 45
4.4 Discussion 47
4.4.1 Main Findings 47
4.4.2 Limitations and Future Directions 49
V Torque-Based Sarcopenia Assessment in Daily Tasks 50
5.1 Background and Related Works 50
5.2 Data Collection and Processing 54
5.2.1 Experimental Apparatus 54
5.2.2 Experimental Protocol and Sarcopenia-Related Clinical Metrics 54
5.3 Methods 55
5.3.1 Joint Torque Computation 55
5.3.2 Data Analysis 56
5.4 Results 59
5.4.1 Participant Group Characteristics 59
5.4.2 Correlation of Torque-Based Metrics with Conventional Clinical Sarcopenia Indicators 59
5.4.3 Torque-Based Discrimination Between Sarcopenic and Non-Sarcopenic Groups 61
5.4.4 Interrelationships among Conventional Clinical Sarcopenia Indicators 64
5.5 Discussion 65
5.5.1 Main Findings 65
5.5.2 Limitations and Future Directions 69
VI Conclusion 71
References 73
Appendix A Data Processing 89
Appendix B Compute GRF in OpenSim 93
Appendix C Coordinate Transformations 94
Appendix D Parallel Data Collection in OpenSim 96
Appendix E Model Configuration 97
Acknowledgements 10
Multimodal Dialogue Act Classification Using Context-Aware Residual Attention and LoRA-Based Cross-Attention Fusion YeJin Park College of Information and Computing Department of AI Convergence Gwangju Institute of Science and Technology
No full textThis paper proposes a multimodal Dialogue Act Classification model that leverages both audio and text modalities to selectively integrate contextual information and perform cross-modal fusion. To preserve utterance-level distinctions while effectively utilizing preceding dialogue, we introduce a Context-Aware Residual Attention mecha- nism that incorporates information from past utterances through residual connections. Furthermore, a Low-rank Gated Cross Attention Fusion module combines multi-head attention with gating mechanisms to model fine-grained interactions between context- aware audio and text embeddings. Evaluated on the EMOTyDA dataset using two pre- ceding utterances as context, the proposed model achieved a macro F1 score of 0.607 and an accuracy of 0.6547, outperforming the baseline by 13.30 and 6.71 percentage points, respectively. Even in settings without contextual input, the model outperformed context-dependent baselines, demonstrating the effectiveness of the proposed context integration and modality fusion strategies.|본 논문은 오디오와 텍스트 정보를 활용하여 문맥 정보를 선택적으로 통합하고, 교차 모달 융합을 수행하는 멀티모달 대화 화행 분류 모델을 제안한다. 제안된 모델은 발화 간 구분성을 유지하면서도 선행 대화로부터 유의미한 정보를 효과적으로 활용하기 위해, 문맥 인지 잔차 어텐션 구조를 도입하여 과거 발화에 대한 주의를 잔차 연결 방식으로 반영한다. 또한, 저랭크 기반 게이트 교차 어텐션 융합 구조는 멀티헤드 어텐션과 게이팅 메커니즘을 결합하여 문맥을 포함한 오디오 및 텍스트 임베딩 간의 정밀한 상호작용을 효과적으로 모델링한다. EMOTyDA 데이터셋을 기반으로 두 개의 이전 발화를 문맥으로 사용할 경우, 제안 모델은 매크로 F1 점수 0.607, 정확도 0.6547을 기록하며 베이스라인 대비 각각 13.30와 6.71 퍼센트포인트 향상된 성능을 보였다. 문맥을 사용하지 않는 설 정에서도 제안 모델은 문맥을 사용하는 기존 베이스라인보다 우수한 성능을 나타내며, 제안한 문맥 통합 및 모달 융합 전략의 효과성을 입증하였다.Master1 Introduction 1
1.1 Overview of Dialogue Act Classification 1
1.2 Motivation 4
1.3 Aim of Thesis 6
2 Baseline Models 8
2.1 Overview 8
2.2 Baseline Model I: Emotion-Aided Multimodal Attention Model (E-MAD) 9
2.3 Baseline Model II: Hierarchical Fusion for Multimodal Dialogue Act
Classification (HF-DAC) 10
3 Proposed Method 13
3.1 Overview 13
3.2 Context-Aware Residual Attention 15
3.3 LoRA-Based Cross-Attention Fusion 17
3.4 Advantages of the Proposed Architecture 18
4 Experiments 20
4.1 Dataset 20
4.2 Evaluation Metric 21
4.3 Experiment Setup 22
4.4 Results 23
4.5 Visualization 27
– iv –
5 Conclusion 30
5.1 Conclusion 30
5.2 Further Study 31
References 33
Acknowledgements 37
– v