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 transferabilty. 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.MasterAbstract i
Contents ii
List of Tables iv
List of Figures vi
Chapter 1. Introduction 1
1.1 Introduction 1
Chapter 2. Related Works 5
2.1 Data-dependent Universal Attack 5
2.2 Data-free Universal Attack 5
2.3 Input Transformation Attack 6
Chapter 3. Data-free Universal Adversarial Perturbation with
Pseudo-semantic Prior 7
3.1 Approach 7
3.1.1 Preliminaries of data-free UAP 7
3.1.2 Pseudo-Semantic Prior 8
3.1.3 Input Transformation 10
3.1.4 Sample Reweighting 11
3.1.5 Overall Loss 12
3.2 Experiments 13
3.2.1 Evaluation on White-Box Attack 14
3.2.2 Evaluation on Black-Box Attack 16
3.2.3 Ablation Study 17
ii
Chapter 4. Conclusion 26
4.1 Conclusion 26
References 27
Summary (in Korean) 32
– iii
3rd Workshop on Maritime Computer Vision (MaCVi) 2025: Challenge Results
No full textThe 3rd Workshop on Maritime Computer Vision (MaCVi) 2025 addresses maritime computer vision for Unmanned Surface Vehicles (USV) and underwater. This report offers a comprehensive overview of the findings from the challenges. We provide both statistical and qualitative analyses, evaluating trends from over 700 submissions. All datasets, evaluation code, and the leaderboard are available to the public at https://macvi.org/workshop/macvi25. © 2025 IEEE
Multi-magnetic phases of FeRh films induced by optical excitation and ion irradiation
No full textIn this dissertation, ultrafast electrodynamics during a unique first-order phase transition between antiferromagnetic (AFM) and ferromagnetic (FM) states transition of metallic compound FeRh are explored using a terahertz time-domain spectroscopy and various time-resolved techniques. Additionally, we investigate the effect of hydrogen ion irradiation on the spatial magnetic distribution and spin precession dynamics in FeRh.
First, we utilized THz time-domain spectroscopy (THz-TDS) and optical pump THz probe (OPTP) techniques to explore how free electron dynamics in FeRh respond to both thermally-driven and photo-induced phase transitions. During the photo-induced phase transition, the transient conductivity exhibits distinct behavior from that of the thermally-driven phase transition as pump fluence surpasses the threshold pump fluence (2.9 mJ/cm²), where the optical pulse induces rapid changes in free electron dynamics driven by photo-excited electrons and notable spin fluctuations. To gain a comprehensive view of the temporal evolution of electronic, lattice, and magnetic dynamics during the photo-induced phase transition, we incorporate time-resolved reflectivity (TR-R) and time-resolved magneto-optical Kerr effect (TR-MOKE) measurements. Furthermore, to understand the fundamental physics of the transition in a non-equilibrium state, we investigated how free electron dynamics respond under various pump fluences around the critical pump fluence for the photo-induced phase transition, with temperature dynamics analyzed based on the three-temperature model (3TM).
Second, we explored the potential of FeRh for creating the FM/AFM multilayer within a single material by exploiting its tunability of the temperature-dependent AFM-FM transition with hydrogen ion (H⁺) irradiation. We investigated bulk and surface magnetic states separately based on the magneto-optical Kerr effect and magnetization-induced second harmonic generation, respectively, and revealed that FeRh can host the FM (surface)/AFM (bulk) magnetic multilayer within a single layer of FeRh even at room temperature, prepared with a H⁺-ion dose of 2.0 × 10¹⁵ H⁺/cm². As the FM and AFM states are stabilized with a well-defined spatial separation as manifested by the exchange bias effect, we expect the FeRh-based FM/AFM bilayer to alleviate limitations arising from the interfaces formed by otherwise different materials.
Finally, we investigate the impact of H⁺-ion irradiation induced defect concentration on spin precession dynamics in FeRh thin films. To elucidate the effects of ion irradiation, we first examine variations in coherent spin precession during the temperature-driven antiferromagnetic (AFM) to ferromagnetic (FM) phase transition in as-grown and H⁺-ion irradiated FeRh samples. Additionally, by comparing spin precession dynamics between laser-induced and temperature-driven phase transitions, we reveal distinct transition pathways resulting from differences in different heat-distributions. Furthermore, field-dependent measurements of spin precession dynamics in the FM state of H⁺-ion irradiated FeRh at room temperature demonstrate the tunability of spin precession dynamics on ion irradiated FeRh films.DoctorAbstract (English) i
List of Contents iv
List of Figures vii
1 Introduction 1
1.1 Phase transitions in condensed matter systems 1
1.2 First order phase transition in metallic compound FeRh 2
1.3 Free electron dynamics in FeRh 4
1.4 Modification of magnetic properties in FeRh through hydrogen ion irra-
diation 8
1.5 Spin precession dynamics in ferromagnetic FeRh films 9
1.6 Outline of thesis 10
2 Experimental techniques 12
2.1 THz time-domain spectroscopy (THz-TDS) 12
2.2 Magneto-optical Kerr effect (MOKE) 13
2.3 Magnetization induced second harmonic generation (MSHG) 15
2.4 Hydrogen ion irradiation on FeRh thin films 16
3 Evolution of electrodynamics upon thermally-driven and photo-induced
phase transition in FeRh 19
3.1 Introduction 19
3.2 Experimental techniques 21
3.2.1 Sample preparation and basic characterization 21
3.2.2 Optical measurements 23
3.3 Results and discussion 25
3.3.1 Temperature dependent complex conductivity spectra for thermally-
driven phase transition 25
3.3.2 Pump fluence dependent OPTP decay profiles for photo-induced
phase transition 29
– iv –
3.3.3 Temporal evolution of free electron dynamics during photo-induced
phase transition 36
3.3.4 Comparison among the results of OPTP, TR-R, and TR-MOKE
measurements 39
3.3.5 Pump fluence dependence of free electron dynamics at non-equilibrium
state 44
3.4 Summary 51
4 Surface and bulk characterization of magnetic multilayers formed
within a single layer FeRh by hydrogen ion irradiation 53
4.1 Introduction 53
4.2 Materials and methods 55
4.2.1 Sample preparation and basic characterization 55
4.2.2 Optical measurements 58
4.3 Results and discussion 59
4.3.1 Magnetic-field dependent MOKE and MSHG results for as-grown
FeRh film during magnetic phase transition 59
4.3.2 Magnetic-field dependent MOKE and MSHG results for hydrogen-
ion irradiated FeRh film during magnetic phase transition 64
4.3.3 Temperature dependent magnetic properties of surface and bulk
for as-grown and hydrogen-ion irradiated FeRh films 67
4.4 Summary 70
5 Laser-induced spin precession in as-grown and hydrogen ion irradi-
ated FeRh films 71
5.1 Introduction 71
5.2 Experimental techniques 72
5.2.1 Sample preparation and basic characterization 72
5.2.2 Optical measurements 75
5.3 Results and discussion 76
5.3.1 Temperature dependent laser-induced spin precession in as-grown
and H+-ion irradiated FeRh films 76
– v –
5.3.2 pump-fluence dependent laser-induced spin precession in weakly
hydrogen ion irradiated FeRh films 80
5.3.3 Magnetic-field dependent laser-induced spin precession in heavily
hydrogen ion irradiated FeRh films 83
5.4 Summary 86
Summary 87
References 89
A X-ray diffraction and reflectivity patterns 107
B Three-temperature model simulations 109
– vi
Hydrophobic Radiative Cooling using Zein-Functionalized Polyvinyl Alcohol Nanofibers with Dielectric Nanoparticles
No full textPolyvinyl alcohol (PVA) is a promising material for radiative cooling owing to its high infrared emissivity and mechanical flexibility; however, its inherent hydrophilicity limits its practical applications, particularly in humid environments. In this study, a scalable and environmentally friendly approach is introduced to overcome this limitation by incorporating Zein-a hydrophobic protein derived from corn-into a PVA matrix in conjunction with aluminum oxide and silicon dioxide nanoparticles. The resulting composite nanofiber membrane, PZAS, exhibits significantly enhanced hydrophobicity, achieving an average water contact angle of 118.5 degrees over 60 s, compared with approximate to 40 degrees for pure PVA nanofibers. A high solar reflectance of 91.7% and strong infrared emissivity of 96.9% within the atmospheric transparency window are demonstrated. In outdoor measurements, a temperature reduction of up to 6.9 degrees C is achieved below ambient temperature. These findings underscore the potential of PZAS as a viable and sustainable radiative cooling material for applications in thermal regulation, particularly in textiles and wearable cooling technologies. The initial hydrophobicity followed by gradual water absorption of PZAS highlights its suitability for advanced biomedical applications, including moisture-managing textiles, sweat-based biosensors, and controlled drug delivery systems.TRUEsciescopu
2D Materials in Logic Technology: Power Efficiency and Scalability in 2DM-MBC CFET
No full textSustaining digital evolution demands high-performance logic technology with a high density, high speed, and low power consumption to process large data sets efficiently. Power consumption remains a critical issue in miniaturized logic devices, impacting reliability, device lifetime, and circuit scalability. This review explores key parameters in logic FETs to manage power consumption, examining advancements in both unit and array structures. We provide a detailed overview of the development history of logic FETs, highlighting structural innovations and challenges for achieving low power consumption. Furthermore, we investigate the state-of-the-art potential of 2D materials (2DMs) in 3D-stacked structures, such as 2DM-MBC CFETs, emphasizing their benefits for ultralow power devices. Finally, we address the current progress and challenges in developing 2DM NMOS and PMOS for CFET industrialization and present an outlook on advancing 2DM-MBC CFET technology to meet the demands of future logic technology. © 2025 American Chemical Society.FALSEsciescopu
Elite Episode Replay Memory for Polyphonic Piano Fingering Estimation
No full textPiano fingering estimation remains a complex problem due to the combinatorial nature of hand movements and no best solution for any situation. A recent model-free reinforcement learning framework for piano fingering modeled each monophonic piece as an environment and demonstrated that value-based methods outperform probability-based approaches. Building on their finding, this paper addresses the more complex polyphonic fingering problem by formulating it as an online model-free reinforcement learning task with a novel training strategy. Thus, we introduce a novel Elite Episode Replay (EER) method to improve learning efficiency by prioritizing high-quality episodes during training. This strategy accelerates early reward acquisition and improves convergence without sacrificing fingering quality. The proposed architecture produces multiple-action outputs for polyphonic settings and is trained using both elite-guided and uniform sampling. Experimental results show that the EER strategy reduces training time per step by 21% and speeds up convergence by 18% while preserving the difficulty level and result of the generated fingerings. An empirical study of elite memory size further highlights its impact on training performance in solving piano fingering estimation.TRUEsciescopu
High-Resolution and Real-Time Interface Detection of Thin Multilayer Films Using Femtosecond Laser-Induced Breakdown Spectroscopy
No full textLaser-Induced Breakdown Spectroscopy (LIBS) is an advanced spectroscopic technique utilized for elemental analysis of materials. The technique involves directing a high-intensity laser pulse onto a material’s surface, inducing rapid energy absorption and generating a microplasma. As this plasma cools, it emits characteristic spectral emissions, which can be analyzed to determine the elemental composition of the material. LIBS is widely employed in various fields, including industrial quality control, environmental monitoring, and biomedical diagnostics, due to its rapid, non-contact nature, minimal sample preparation, and real-time analytical capabilities.
However, conventional LIBS is inherently limited by spatial resolution and signal stability, particularly when applied to precision laser processing of thin film layers. Femtosecond LIBS (fs-LIBS) has emerged as a high-resolution alternative, significantly reducing thermal damage and enhancing spectral accuracy. This study explores the application of high-resolution fs-LIBS for real-time interface detection during the laser processing of thin film layers. In prior studies on interface detection using laser-induced plasma spectroscopy, there have been no reports of successful plasma signal acquisition or interface identification under laser processing conditions with a lateral resolution below 10 μm. This limitation is primarily attributed to the intrinsic difficulty of generating stable and analyzable plasma emissions when the ablation volume is extremely small. As the lateral resolution increases, the resulting plasma becomes highly confined and unstable, significantly hindering the reliability of emission-based detection methods. In the present study, stable plasma signals were successfully acquired using laser-induced craters with a lateral resolution of approximately 3 μm and a depth resolution of around 400 nm. This result demonstrates that the proposed method is highly applicable to the precision processing of multilayer thin films, particularly in applications where feature sizes are on the order of a few micrometers, such as in semiconductor and display panel manufacturing. Accurate detection of layer transitions within multi-layered structures is critical for advanced manufacturing processes, including semiconductor fabrication and optical coatings. Traditional interface detection techniques often lack the necessary spatial and temporal resolution for precise material removal, leading to inefficiencies and defects. In contrast, fs-LIBS facilitates real-time monitoring of laser-material interactions, enabling precise control over depth profiling and interface identification.
The research systematically examines the influence of key laser parameters—including pulse duration, fluence, and repetition rate—on the accuracy of interface detection. A series of experiments were conducted using femtosecond laser pulses to process thin film layers while simultaneously capturing emission spectra to determine layer composition and thickness.
Furthermore, this study introduces a novel real-time interface detection method based on high-resolution fs-LIBS. Unlike previous approaches relying on the crossover point of signal intensities between upper and lower layers, our method defines the interface as the point where the emission signal of the lower layer surpasses the noise threshold. To quantitatively identify noise regions, various statistical techniques—including signal classification, interquartile range (IQR), and confidence interval analysis—were employed. These methods enabled the effective separation of true signal from background noise and allowed for the reliable determination of signal occurrence. As a result, even under low pulse energy conditions with weak emission intensity, the interface could be detected without signal averaging, thereby demonstrating strong applicability to real-time processing environments.
The findings demonstrate that fs-LIBS significantly enhances the resolution and accuracy of interface detection compared to conventional laser processing methods. The ability to detect layer transitions in real time improves process control, minimizes defects, and increases manufacturing efficiency. The results further underscore the potential of fs-LIBS in high-precision applications, such as microelectronics, aerospace, and biomedical engineering, where thin film processing plays an essential role.
Ultimately, this research presents a novel approach to integrating high-resolution laser processing with real-time analytical techniques. By leveraging fs-LIBS, manufacturers can achieve superior accuracy in multi-layered material processing, contributing to advancements in next-generation fabrication technologies.DoctorAbstract i
국문 초록 iv
Contents vi
List of Tables viii
List of Figures ix
1 Introduction 1
1.1 Research background 1
1.2 Interface prediciton methods based on the LIBS 3
1.2.1 Conventional depth profiling methods using LIBS 3
1.2.2 Limitations of intersection-based interface detection 4
1.3 Research objective 7
2 Study on effect of laser energy conditions on multilayer thin film in LIBS 9
2.1 Experiment 9
2.1.1 Sample 9
2.1.2 LIBS setup 10
2.2 Results and discussion 14
2.2.1 Resolution dependence on laser pulse energy 14
2.2.2 Background Subtraction 17
2.2.3 Shot to shot LIBS intensity fluctuation 18
2.2.4 SEM image-based morphological abalysis 19
2.3 Summary 25
3 Interface detection algorithm process 26
3.1 Noise level analysis 26
3.1.1 The types of signal peaks 26
3.1.2 The process of removing signal peaks 28
3.2 Interface decision procedure 32
3.2.1 Confidence interval 32
3.2.2 Blind test 33
3.3 Verificaiton 39
3.3.1 Algorithm results 39
3.3.2 Cross-section analysis 39
3.4 Summary 42
4 Conclusions 43
References 45
Acknowledgement 5
Structural insights into enzymatic mechanisms in the biosynthetic pathway of phosphatidylethanolamine
No full textPhospholipids are essential components of cellular membranes, playing critical roles in maintaining membrane integrity, mediating signal transduction, and regulating homeostasis. This thesis investigates conserved and divergent features of phospholipid biosynthesis across prokaryotic and eukaryotic systems, with a focus on two key enzymes. The introduction outlines the fundamental roles of phospholipids and summarizes membrane- associated biosynthetic pathways in bacteria and mammals, highlighting distinct regulatory strategies. Chapter I focuses on Escherichia coli phosphatidylserine synthase (PssA), which catalyzes the formation of phosphatidylserine from CDP-diacylglycerol and L-serine. Crystal structures of PssA in substrate- free and substrate-bound states uncover the molecular basis for substrate recognition and covalent intermediate formation. Sedimentation velocity experiments show that cytosolic PssA exists in a monomer-dimer equilibrium, with only the monomer capable of membrane association, suggesting an oligomerization-dependent regulatory mechanism. Chapter II presents the cryo-electron microscopy structure of mouse PCYT2α and β, the rate-limiting enzyme in the Kennedy pathway for phosphatidylethanolamine synthesis. PCYT2 forms a symmetric homodimer, with the N-terminal cytidylyltransferase (CT) domain mediating catalysis via a conserved HXGH motif, while the C-terminal CMP-binding domain likely plays a non-catalytic structural or regulatory role. Mutational and kinetic analyses support this functional asymmetry. Together, these studies reveal both shared and distinct features of phospholipid biosynthetic enzymes in bacteria and mammals, improve our understanding of lipid metabolism, and provide structural information that may help guide the development of antibiotics and treatments for metabolic diseases.|인지질은 세포막의 필수 구성 요소로, 막의 구조적 안정성 유지, 신호 전달 조절, 항상성 유지 등
다양한 생물학적 기능을 수행한다. 본 학위논문에서는 원핵생물과 진핵생물에서의 인지질 생합성
경로의 보존된 특징과 상이한 조절 메커니즘을 비교 분석하였으며, 이 과정에서 핵심적인 역할을
하는 두 효소를 중심으로 구조 및 기능을 규명하였다.
서론에서는 인지질의 기본적인 역할을 서술하고, 세균과 포유류에서의 막 관련 인지질 생합성
경로를 요약하였다. 이를 통해 각 생물 군에서 채택한 상이한 조절 전략을 강조하였다.
제 1 장은 Escherichia coli 의 phosphatidylserine synthase (PssA)를 다룬다. PssA 는 CDPdiacylglycerol 과 L-serine 을 기질로 하여 phosphatidylserine 을 합성하는 효소이다. 기질이
결합되지 않은 상태와 결합된 상태의 결정 구조를 통해, 기질 인식 및 공유 중간체 형성에
관여하는 분자적 기전을 규명하였다. 침강 속도 분석 결과, 세포질 내 PssA 는 단량체와 이량체
사이의 평형 상태에 있으며, 막 결합 능력은 단량체에만 존재함을 확인하였다. 이는 효소의
올리고머화 상태에 따른 조절 메커니즘이 존재함을 시사한다.
제 2 장에서는 phosphatidylethanolamine 생합성 경로(Kennedy 경로)의 속도 결정 효소인 생쥐
유래 PCYT2α 및 β의 단일입자 초저온 전자현미경 구조를 제시한다. PCYT2는 대칭적인 이량체를
형성하며, N-말단의 cytidylyltransferase(CT) 도메인이 보존된 HXGH 모티프를 통해 촉매 활성을
수행하고, C-말단 CMP 결합 도메인은 구조적 또는 조절적 역할을 수행하는 것으로 추정된다.
돌연변이 분석 및 효소 반응 속도 측정을 통해 이러한 도메인 간 기능 비대칭성이 뒷받침되었다.
이와 같은 연구 결과는 박테리아와 포유류 인지질 생합성 효소 간의 공통점과 차이점을 규명하며,
지질 대사에 대한 이해를 심화시키고, 항생제 및 대사 질환 치료제 개발을 위한 구조 기반 정보를
제공할 수 있을 것으로 기대된다.DoctorIntroduction Phospholipid biosynthesis and regulation in membrane biology 1
I.1. Phospholipids: central components of cellular function 2
I.2. Phospholipid biosynthesis 4
I.3. Membrane-associated enzymes and their regulation 9
Chapter I Structural and functional analysis of phosphatidylserine synthase (PssA) in bacteria 11
1.1. Abstract 12
1.2. Introduction 13
1.3 Material and Methods 18
1.4. Results 26
1.5. Discussion 69
Chapter II Structural and functional analysis of mammalian CTP:phosphoethanolamine cytidylyltransferase (PCYT2) 74
2.1. Abstract 75
2.2. Introduction 76
2.3. Material and methods 85
2.4. Results 92
2.5. Discussion 120
Reference 130
Curriculum Vitae 139
Acknowledgements 141
국문 초록 14
Discovery of lanthanide metal oxide catalyst for transesterification reaction by fluorescence-based high-throughput screening method and application to biodiesel production
No full textThe development of heterogeneous metal oxide catalysts for transesterification reactions is crucial owing to their seamless reusability and environmental friendliness. In recent years, numerous studies have been conducted on rare-earth oxides, such as lanthanide metal oxides. Various metal oxides were screened for transesterification using a new fluorescence-based high-throughput screening (HTS) method with a pyrene excimer probe, bis(4-(1-pyrenyl)butyl) maleate (BPBM). Praseodymium(iv) oxide (PrO2) yielded the highest catalytic activity among the prepared metal oxides. Various substrates were successfully transesterified, and biodiesel was produced in a high yield (90%) from soybean oil through transesterification using the catalyst. The selected catalyst required minimal amounts for the transesterification of various organic substrates (0.7 mol%) and soybean oil (0.8 wt%). © 2025 The Royal Society of Chemistry.TRUEsciescopu
DD-PRiSM: a deep learning framework for decomposition and prediction of synergistic drug combinations
No full textCombination therapies have emerged as a promising approach for treating complex diseases, particularly cancer. However, predicting the efficacy and safety profiles of these therapies remains a significant challenge, primarily because of the complex interactions among drugs and their wide-ranging effects. To address this issue, we introduce DD-PRiSM (Decomposition of Drug-Pair Response into Synergy and Monotherapy effect), a deep-learning pipeline that predicts the effects of combination therapy. DD-PRiSM consists of two predictive models. The first is the Monotherapy model, which predicts parameters of the drug response curve based on drug structure and cell line gene expression. This reconstructed curve is then used to predict cell viability at the given drug dosage. The second is the Combination therapy model, which predicts the efficacy of drug combinations by analyzing individual drug effects and their synergistic interactions with a specific dosage level of individual drugs. The efficacy of DD-PRiSM is demonstrated through its performance metrics, achieving a root mean square error of 0.0854, a Pearson correlation coefficient of 0.9063, and an R2 of 0.8209 for unseen pairs. Furthermore, DD-PRiSM distinguishes itself by its capability to decompose combination therapy efficacy, successfully identifying synergistic drug pairs. We demonstrated synergistic responses vary across cancer types and identified hub drugs that trigger synergistic effects. Finally, we suggested a promising drug pair through our case study. © The Author(s) 2025.TRUEsciescopu