Daegu Gyeongbuk Institute of Science and Technology
DGIST Library Institutional RepositoryNot a member yet
12664 research outputs found
Sort by
NDPipe: Exploiting Near-data Processing for Scalable Inference and Continuous Training in Photo Storage
No full textThis paper proposes a novel photo storage system called NDPipe, which accelerates the performance of training and inference for image data by leveraging near-data processing in photo storage servers. NDPipe distributes storage servers with inexpensive commodity GPUs in a data center and uses their collective intelligence to perform inference and training near image data. By efficiently partitioning deep neural network (DNN) models and exploiting the data parallelism of many storage servers, NDPipe can achieve high training throughput with low synchronization costs. NDPipe optimizes the near-data processing engine to maximally utilize system components in each storage server. Our results show that, given the same energy budget, NDPipe exhibits 1.39× higher inference throughput and 2.64× faster training speed than typical photo storage systems. © 2024 Copyright held by the owner/author(s)
Specification of neural circuit architecture shaped by context-dependent patterned LAR-RPTP microexons
Get PDFLAR-RPTPs are evolutionarily conserved presynaptic cell-adhesion molecules that orchestrate multifarious synaptic adhesion pathways. Extensive alternative splicing of LAR-RPTP mRNAs may produce innumerable LAR-RPTP isoforms that act as regulatory “codes” for determining the identity and strength of specific synapse signaling. However, no direct evidence for this hypothesis exists. Here, using targeted RNA sequencing, we detected LAR-RPTP mRNAs in diverse cell types across adult male mouse brain areas. We found pronounced cell-type–specific patterns of two microexons, meA and meB, in Ptprd mRNAs. Moreover, diverse neural circuits targeting the same neuronal populations were dictated by the expression of different Ptprd variants with distinct inclusion patterns of microexons. Furthermore, conditional ablation of Ptprd meA+ variants at presynaptic loci of distinct hippocampal circuits impaired distinct modes of synaptic transmission and objection-location memory. Activity-triggered alterations of the presynaptic Ptprd meA code in subicular neurons mediates NMDA receptor-mediated postsynaptic responses in CA1 neurons and objection-location memory. Our data provide the evidence of cell-type- and/or circuit-specific expression patterns in vivo and physiological functions of LAR-RPTP microexons that are dynamically regulated. © The Author(s) 2024.TRUEsciescopu
Fabrication of silicon-based nickel nanoflower-encapsulated gelatin microspheres as an active antimicrobial carrier
No full textLocal antibiotic application might mitigate the burgeoning problem of rapid emergence of antibiotic resistance in pathogenic microbes. To accomplish this, delivery systems must be engineered. Hydrogels have a wide range of physicochemical properties and can mimic the extracellular matrix, rendering them promising materials for local antibacterial agent application. Here, we synthesized antibacterial silicon (Si)-based nickel (Ni) nanoflowers (Si@Ni) and encapsulated them in gelatin methacryloyl (GelMA) using microfluidic and photo-crosslink technology, constructing uniform micro-sized hydrogel spheres (Si@Ni-GelMA). Si@Ni and Si@Ni-GelMA were characterized using X-ray diffraction, transmission electron microscopy, and scanning electron microscopy. Injectable Si@Ni-GelMA exhibited excellent antibacterial activities owing to the antibiotic effect of Ni against Pseudomonas aeruginosa, Klebsiella pneumoniae, and methicillin-resistant Staphylococcus aureus, while showing negligible cytotoxicity. Therefore, the Si@Ni-GelMA system can be used as drug carriers owing to their injectability, visible light-mediated crosslinking, degradation, biosafety, and superior antibacterial properties. © 2024FALSEsciescopu
AN APPARATUS FOR CONTROLLING THE MOVEMENT OF PARTICLES Bz USING A CONTROL OF A MAGNETIC FIELD, AND A METHOD FOR CONTROLLING THE MOVEMENT OF PARTICLES USING THE SAME
No full text본 발명의 일 실시 예는 자기장 제어를 이용한 입자 이동 제어장치 및 이를 이용한 입자 이동 제어방법을 제공한다. 본 발명의 실시 예에 따른 자기장 제어를 이용한 입자 이동 제어장치는, 상부를 통과하는 미세입자를 향해 자기장 영역이 일정하게 제공되는 자기장인 고정자기장을 발생시키는 고정자기장발생부; 및 고정자기장발생부의 상단에 형성되고, 자기장 영역이 가변되는 자기장인 조작자기장을 발생시켜 미세입자에 제공하는 조작자기장발생부를 포함한다
Hydrogen-Driven Low-Temperature Topotactic Transition in Nanocomb Cobaltite for Ultralow Power Ionic-Magnetic Coupled Applications
No full textWe reversibly control ferromagnetic-antiferromagnetic ordering in an insulating ground state by annealing tensile-strained LaCoO3 films in hydrogen. This ionic-magnetic coupling occurs due to the hydrogen-driven topotactic transition between perovskite LaCoO3 and brownmillerite La2Co2O5 at a lower temperature (125-200 °C) and within a shorter time (3-10 min) than the oxygen-driven effect (500 °C, tens of hours). The X-ray and optical spectroscopic analyses reveal that the transition results from hydrogen-driven filling of correlated electrons in the Co 3d-orbitals, which successively releases oxygen by destabilizing the CoO6 octahedra into CoO4 tetrahedra. The transition is accelerated by surface exchange, diffusion of hydrogen in and oxygen out through atomically ordered oxygen vacancy “nanocomb” stripes in the tensile-strained LaCoO3 films. Our ionic-magnetic coupling with fast operation, good reproducibility, and long-term stability is a proof-of-principle demonstration of high-performance ultralow power magnetic switching devices for sensors, energy, and artificial intelligence applications, which are keys for attaining carbon neutrality. © 2024 American Chemical Society.FALSEsciescopu
Poly-Melamine Coated Copper Nanowires for Efficient Electrochemical CO2 Reduction
No full textCopper nanowire; Poly-melamine; Core-shell nanostructure; CO2 reduction reactionAbstract i
List of contents ii
List of tables iii
List of figures vi
Ⅰ. Introduction 1
1.1 Copper nanowire for CO2 reduction 1
1.2 Poly-melamine coated CuNW 2
1.3 Influences of chemical environment on CO2 reduction 3
1.4 Influences of porous shell on CO2 reduction 4
Ⅱ. Theoretical background 5
2.1 Cpper nanowire (CuNW) 5
2.1.1 Growth mechanism of CuNWs 5
2.1.2 Synthesis methods for CuNWs 7
2.1.3 Purification of CuNWs 8
2.2 Synthesis methods for polymer coated metal nanowires 10
2.2.1 Electropolymerization 10
2.2.2 In situ polymerization 12
2.3 Electrochemical CO2 redcution reaction (CO2RR) 13
2.3.1 Copper for electrochemical CO2RR 15
2.3.2 The mechanism of CO2RR on Cu Surface 17
2.3.3 Modifier addition to copper for enhanced CO2RR 19
2.4 The characteristics of melamine 25
2.4.1 High interaction with CO2 25
2.4.2 Abundant and ordered reaction sites 27
Ⅲ. Experimental Section 30
3.1 Chemicals and materials 30
3.2 Synthesis of CuNWs 30
3.3 Preparation of poly melamine coated CuNWs 30
3.4 Preparation of poly-melamine 31
3.5 Characterization 31
3.6 Electrode preparation and electrochemical CO2RR measurement 32
Ⅳ. Result and discussion 35
4.1 Poly-melamine coated CuNWs synthesis 35
4.1.1 Synthesis of poly-melamines 35
4.1.2 Stirring method 36
4.1.3 Hydrothermal method 41
4.1.4 Reflux heating method 43
4.2 Poly-melamine as a modifier for catalyst activity enhancement 48
4.2.1 Influence of poly-melamine chemical composition on catalyst activity 50
4.2.2 Influence of poly-melamine pore on catalyst activity 54
Ⅴ. conclusion 58
Reference 59
Summary 64MasterdCollectio
동형 암호를 활용한 민감 정보 보호 아래 기반 시설 이상 탐지 시스템 구축
No full textCritical Infrastructure; 기반 시설;Anomaly Detection System; 이상 탐지 시스템;Homomorphic Encryption; 동형 암호Critical infrastructure (CI) has been vulnerable to various threats due to its crucial role in society. While there are several general methods for detecting anomalies in time series data, such as long short-term memory (LSTM) and Transformer, these may not be available on local computing resources at CI. With the advent of cloud computing, users can use computing resources on demand to utilize various models. However, cloud computing has a shortcoming in privacy since a user must send the data to a cloud server for processing. Such a situation is critical, especially for CI, which deals with sensitive data. While data can be stored securely by encrypting it first and transmitting it to a cloud server, encrypted data cannot generally undergo arbitrary operations. Fully homomorphic encryption (FHE) is a scheme that allows arbitrary numbers of operations on encrypted data. Applying homomorphic encryption allows the construction of a privacy-preserving anomaly detection model in a cloud computing environment. However, FHE requires a computationally expensive operation called bootstrapping on every certain number of multiplications to refresh encrypted data. Therefore, to use homomorphic encryption practically, minimizing the number of multiplications is necessary to avoid bootstrapping. This work will propose a lightweight anomaly detection model by utilizing the properties of critical infrastructure. The model consists of multiple sub-models which detect anomalies under a specific constant subsystem state. Training the model with encrypted data is verified in the SWaT dataset and showed competitive performance compared to another simple approach available under a homomorphic encryption scheme.|기반 시설이 사회 전반에서 주요한 역할을 함에 따라 외부로부터의 공격 대상이 되는 일이 늘어나고 있다. 일반적인 시계열 데이터를 대상으로 하는 이상 탐지 시스템이 여러 알려져 있지만 기반 시설 내의 컴퓨팅 자원으로는 이러한 시스템을 활용하기 어려울 수 있다. 클라우드 컴퓨팅 환경이 등장하면서 사용자는 개인 컴퓨팅 자원에 제한되지 않고 필요한 만큼 컴퓨터 자원을 사용할 수 있게 되었으나, 데이터를 연산하기 위해서는 그 데이터를 클라우드 서버에 전송해야 하는 보안적인 약점이 존재한다. 이는 민감한 정보를 다루는 기반 시설에 더욱 문제를 야기할 수 있다. 클라우드에 데이터를 전송할 때 앞서 데이터를 암호화하여 저장하면 데이터 내의 정보가 유출되지는 않지만 일반적으로 암호화된 데이터는 연산이 불가능하다. 하지만 동형 암호로 암호화된 데이터는 덧셈과 곱셈을 비롯한 연산을 할 수 있어 기반 시설 데이터에 동형 암호를 적용한다면 클라우드 상에서 민감한 정보의 유출 없이 이상 탐지 시스템을 구축할 수 있다. 하지만 동형 암호는 특정 횟수의 곱셈이 수행될 때마다 부트스트래핑이라 불리는 시간이 오래 걸리는 연산을 수행해줘야 한다. 따라서 동형 암호를 효과적으로 사용하기 위해서는 연산 횟수를 최소화하여 부트스트래핑 연산 없이 이상 탐지 모델이 작동하도록 모델을 구축하여야 한다. 본 논문에서는 기반 시설의 속성을 활용하여 경량화된 이상 탐지 모델을 제안한다. 이 모델은 여러 개의 하위 모델로 구성되며 각 하위 모델은 특정 하위 시스템 상황 아래 이상 탐지를 실행한다. 제안하는 모델은 스왓 데이터 세트에서 암호화된 데이터로 학습하는 것이 가능함을 확인하였으며 동형 암호가 적용된 환경 아래 가능한 다른 간단한 방법과 비교해 경쟁력 있는 성능을 보여주었다.List of Contents
Abstract i
List of contents ii
List of tables · vi
List of figures · v
Ⅰ. Introduction 1
1.1. Problem Statement 2
Ⅱ. Background 3
2.1. Homomorphic Encryption 3
2.1.1. Categories of Homomorphic Encryption Schemes 3
2.1.2. Examples of Fully Homomorphic Encryption Schemes 4
Ⅲ. Related Work 6
3.1. Integrating Machine Learning and Homomorphic Encryption 6
3.2. Anomaly Detection on Time Series Data 6
Ⅳ. Privacy-Preserving Anomaly Detection Model · 8
4.1. The Model Design 8
4.2. Training the Model with Encrypted Data 10
Ⅴ. Experiments 12
5.1. SWaT Dataset 12
5.2. Data Preprocessing 13
5.3. Results 15
Ⅵ. Discussion · 17
Ⅶ. Conclusion 18MasterdCollectio
From Homogeneity to Turing Pattern: Kinetically Controlled Self-Organization of Transmembrane Protein
No full textUnderstanding the spatial organization of membrane proteins is crucial for unraveling key principles in cell biology. The reaction-diffusion model is commonly used to understand biochemical patterning; however, applying reaction-diffusion models to subcellular phenomena is challenging because of the difficulty in measuring protein diffusivity and interaction kinetics in the living cell. In this work, we investigated the self-organization of the plasmalemma vesicle-associated protein (PLVAP), which creates regular arrangements of fenestrated ultrastructures, using single-molecule tracking. We demonstrated that the spatial organization of the ultrastructures is associated with a decrease in the association rate by actin destabilization. We also constructed a reaction-diffusion model that accurately generates a hexagonal array with the same 130 nm spacing as the actual scale and informs the stoichiometry of the ultrastructure, which can be discerned only through electron microscopy. Through this study, we integrated single-molecule experiments and reaction-diffusion modeling to surpass the limitations of static imaging tools and proposed emergent properties of the PLVAP ultrastructure. © 2024 American Chemical SocietyFALSEsciescopu
TARGET LOCATION DETERMINE APPARATUS USING EXTRAPOLATION AND METHOD THEREOF
No full text타켓 위치 결정 장치가 개시된다. 본 타겟 위치 결정 장치는 레이더 신호를 방출하고, 타겟에 반사된 수신 신호를 수신하는 레이더 장치, 및 수신된 수신 신호에 포함된 복수의 첩(chirp) 신호를 분리하고, 복수의 첩 신호를 이용하여 확장된 수신 신호를 생성하고, 확장된 수신 신호를 이용하여 타겟과 레이더 간의 거리 정보를 생성하는 프로세서를 포함한다