Ordinary Differential Equation Multi-Domain Models of MEMS Structures

Childs, Carter J The Pennsylvania State University ProQuest Dissert 2023 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Modeling is a pivotal part of any modern design process. The ability to accurately predict designs before building cuts down on development costs leading to faster development cycles. Modeling of transducers has been accomplished in various methods, mostly being compartmentalized into mechanical, electrical, or system models. While each of these models is important, the current solutions require substantial interaction between model types. There exists headroom to expand the system-level models to be less dependent on the subsystem model responses. To accomplish this, techniques developed before comprehensive Finite Element Analysis tools are leveraged and updated to work within modern computational tools, while expanding the repertoire to include modern production techniques of Micro-Electrical-Mechanical-Systems (MEMS). The invention of transmission line models that model the response of physical systems commonly found in MEMS sensory devices was undertaken. The models were built in Mathworks Simscape multidomain modeling software, taking advantage of the multiple domain definitions. The models developed can be used to allow system models to work independently of other models by breaking the dependency of system models on the results of subsystem models. Further, the advantages of multidomain models open up the doors to modeling phenomena that are extremely complex and computationally expensive to complete with available tools. New component models were developed for thin bars and plates as well as thermoviscous models of acoustic enclosures.

파라미트릭 점프 프로세스를 위한 최적의 모수 추정 방법론 : Optimal Calibration for Parametric Jump Processes

양승호 포항공과대학교 일반대학원 2011 국내박사

According to numerous empirical evidences observed in option markets, it is clear that the celebrating Black-Scholes-Merton option pricing model can not explain the intrinsic properties of option prices in real markets such as the implied volatility smile behavior. To capture the smile effect many option pricing models or methods have been developed in a non-parametric and parametric way. In non-parametric approaches they do not rely on pre-assumed models but instead try to uncover/induce the model. There is a weak point with non-parametric approaches which it cannot applied to pricing path-dependent exotic options due to its lack of underlying dynamics. Recently in financial literature parametric methods, such as exponential L´evy models and affine jump-diffusion models, have been widely adopted as alternative models that explain stylized facts of asset returns and volatility smile effects of traded option prices. Hence if the parameters are calibrated reasonably the parametric models can be very powerful. Unfortunately the number of parameters is a lot and it’s hard to estimate parameters from the information in financial market. To calibrate them we use cross-sectional data of option prices. Least-square sense is usually employed to calibrate them in finance, although it is well-known ill-posed inverse problem. To conquer the ill-posed inverse problem we propose a derivative-free calibration method constrained by four observable statistical moments (mean, variance, skewness and kurtosis) from underlying time series and so-called multi-basin system which consists of three sequential phases to expedite the search for a good parameter set. To verify the performance of the proposed methods, we conduct simulations on some model-generated option prices data and real-world option market data. The simulation results show that the proposed methods fit the option ranges well and calibrate the parameter set of exponential L´evy models and affine jump-diffusion models reasonably and robustly. In this thesis we also give a modularized summary of all the detailed equations relevant to all exponential L´evy models and affine jumpdiffusion models in a consistent way by using the unified notations.

Gravitational Lensing : Models and Astrophysical Applications

채규현 University of Pittsburgh 1999 해외박사

This dissertation is an investigation of gravitational lens models and detailed analyses of the Tensing properties of three astrophysically interesting multiply-imaged quasi-stellar objects (QSOs) Q2237+0305, Q0957+561, and H1413+1143. A general-form power-law elliptical mass distribution is considered as a model for a Tensing object, and a Fourier series expansion technique is applied to obtain semi-analytical expressions for the deflection and magnification due to the general power-law mass model. Investigating more sophisticated lens models permitted by most recent and comprehensive sets of data, the following astrophysical/cosmological results are derived. For Q2237+0305, Tensing properties of the elliptical mass model with varying radial index are calculated, and similarities and quantitative differences between the elliptical lens model and a pseudo-elliptical lens model (a power-law sphere plus shear) are highlighted. For Q0957+561, the elliptical mass model and a power-law sphere are used to model the Tensing galaxy and cluster respectively, as motivated by recent Hubble Space Telescope (HST) imaging of the galaxy and weak Tensing observations of the cluster. The lens model for Q0957+561 gives a best-fit to present observational constraints and yields a Hubble constant of H_(o) = 57^(+19)_(-16) (95% confidence) km s^(-1) Mpc^(-1). For H1413+1143, recent data from HST WFPC1,2, NICMOS, and FOS are used to explore possible lens models. Analyses of two-component lens models (namely, two-galaxy and galaxy+cluster) indicate that at least the continuum source region is microlensed. The broad emission line region could also be microlensed. However, information from present observations are not sufficient to draw any conclusions on the leasing nature of the broad emission line region or its size scale.

BAYESIAN MULTIVARIATE SPATIAL MODELS AND THEIR APPLICATIONS

송준진 Texas A&M University 2004 해외박사

단변량 계층 베이즈 모델들이 많은 분야에서 활발히 연구되고 있다. 일반적으로 공간 데이터는 다변량인 경우가 많다. 예를 들어, 여러 질병들의 발병들은 일반적으로 공중 위생학에서 카운티나 인구조사 단위로 기록되어 진다. 이와 같은 경우 주어진 공간 단위에서의 질병간의 상관관계 뿐만 아니라 공간적인 상관관계도 고려되어 지는 것이 타당하다. 본 논문에서 지역 단위들에서 수집된 다변량 관측치들을 위한 다변량 공간 모형화를 연구한다. 본 논문에서는 다변량 공간 데이터를 위한 계층 베이즈 모델들을 제안한다. 본 논문에서 텍사스의 카운티 수준의 교통사고 데이터들이 제안된 방법을 설명하기 위해 사용되어 진다. 잠재적인 확장으로서 네트워크 기반의 위험지도를 만들기 위한 단변량 계층 베이즈 모델들의 사용도 논의되어 진다. 여러 종류의 교통사고률을 동시에 추정하기 위해 여러 베이지안 다변량 공간 모델들이 제안된다. 단변량조건자기회기모형군이 공간 모델을 위해 고려되어 진다. 단변량조건자기회귀모형이 다변량 공간 데이터 분석을 위해 일반화 된다. 베이지안 추론에서 부적절사전분포의 사용에서 사후분포의 적절성을 확보하기 위한 정리들이 제안된다. 제안된 모델들의 비교를 위해 편차 정보기준을 사용되어 진다. 마프코프 연쇄 몬테칼로 계산 방법이 모델 모수 추정과 통계적 추론을 위해 사용되어 진다. 마지막으로 본 논문은 간단한 연구 정리와 후속 연구들을 제시하였다. Univariate hierarchical Bayes models are being vigorously researched for use in disease mapping, engineering, geology, and ecology. This dissertation shows how the models can also be used to build model-based risk maps for area-based roadway traffic crashes. County-level vehicle crash records and roadway data from Texas are used to illustrate the method. A potential extension that uses univariate hierarchical models to develop network-based risk maps is also discussed. Several Bayesian multivariate spatial models for estimating the traffic crash rates from different types of crashes simultaneously are then developed. The specific class of spatial models considered is conditional autoregressive (CAR) model. The univariate CAR model is generalized for several multivariate cases. A general theorem for each case is provided to ensure that the posterior distribution is proper under improper and flat prior. The performance of various multivariate spatial models is compared using a Bayesian information criterion. The Markov chain Monte Carlo (MCMC) computational techniques are used for the model parameter estimation and statistical inference. These models are illustrated and compared again with the Texas crash data. There are many directions in which this study can be extended. This dissertation concludes with a short summary of this research and recommends several promising extensions.

An Analysis of UK Urban Fringe Management Models Applying the Actor Network Theory

김용범 Cardiff University 2002 해외박사

The main objectives of this research are to investigate and analyse the operation processes of urban fringe management models using the actor network theory concept, especially the sociology of translation. The perceptions, actions and reactions of local interest groups and agencies to local circumstances produced a complex framework from which particular management models emerged. Imporlantly, because of their limited statutory and financial powers, local authorities were unable to effectively and efficiently deal with localities as a result of the deregulation of property rights of a variety of landownerships. This inability crucially led to the formation of the management models with the anticipation of building a coalition of interest groups and public and non-government organisations in the management processes in order to improve the physical, economic and social environments and facilitate the management mechanism. However, the instigators of the management models did not seek to make themselves indispensable, rather they sought to make their proposals, the management models, indispensable as instruments for constructing actor networks, sustaining and enhancing the environment and managing the urban fringe. The management models' objectives were to gain and maintain the interests of other actors and to ensure their enrolment in their actor networks. Accordingly, the operational processes of the management models were dependent on capturing the interests of other actors. In addition, although the role of speaking on behalf of etch management model's actor network is differently invested in certain internal organisation bodies, such as the governance body and management body, what is common to all the urban fringe management models is that the governance body within each management model tries to speak with one voice to bring about an effective operation process, and the management team employs a variety of devices to secure the enrolment of a variety of actors in its actor network.

Estimating Korean Pine(Pinus koraiensis) Habitat Distribution Considering Climate Change Uncertainty : Using Species Distribution Models and RCP Scenarios = 불확실성을 고려한 잣나무의 서식 적지 분포 예측 : 종 분포 모형과 RCP시나리오를 중심으로

안윤정 서울대학교 대학원 2015 국내석사

Climate change can significantly affect tree species distribution in forests. Therefore, adaptation planning is needed to obtain maximum returns on tree growth. Pinus koraiensis, the common name is Korean pine, is a major afforestation species in Korea and is normally distributed in frigid zones. For this reason, global warming could affect the distribution of the Korean pine. Therefore, this study aimed to predict the distribution of the Korean pine and its suitable habitat area considering uncertainty by applying climate change scenarios in an ensemble model. Species distribution points and environmental variables data were used for the input data in the model. First, a site index was considered when selecting present and absent points by using the stratified method. Secondly, environmental and climate variables were chosen by literature review and then correlation analysis was performed to select variables that were not correlated. Subsequently, the selected variables were confirmed with experts. Those variables were then used as input data of BIOMOD2 (BIOdiversity MODelling 2). Next, the present distribution model was made and the result was validated with data splitting and Receiver Operating Characteristic (ROC). Next, Representative Concentration Pathways (RCPs) scenarios were applied to the models to create the future distribution model. Finally, the ensemble models were built and consensus maps were created using model committee averaging (MCA). In addition, overlay maps and uncertainty maps were used to quantify the uncertainties of the results. The estimated results of the individual models showed significant variation. Among the eight models, Random Forest (RF) had the highest accuracy. The Artificial Neural Network (ANN) model tended to overestimate results, and the Maximum Entropy Algorithm (Maxent) results were distinct from those of the other models. These differences can be explained by the algorithms of each model, the interaction of input data, and the verification methodology. The uncertain area from individual models was excluded from the ensemble model results. In the midterm future (2040s), the models themselves created the major differences observed in Korean pine distribution. In contrast, both the models and RCPs scenarios caused variation in the long-term future (2090s). Results of ensemble models were calculated using uncertainty and overlay maps, with the uncertainty of one overlay map close to 17%. The uncertainty of the five times overlaid area was around 8% in both the midterm and long-term futures. Suitable habitat for the Korean pine in the midterm future is mainly distributed in the central part of Korea, Gangwon province, and the southern part of Korea. In the long-term future, this preferred area will disappear from the southern part of Korea as well as some areas of Gangwon province. Generally, most model and ensemble results predicted that the suitable habitat area would decrease in the mid- and long-term future. As the Korean pine is an afforestation species, it cannot be planted in protected areas. Therefore, protected areas were eliminated from the results of the ensemble model. The ratios of protected area were 25%, 25%, 19%, and 22% in RCPs 2.6, 4.5, 6.0, and 8.5, respectively, in the midterm future. There was no significant difference among the results. The protected area ratios were 24%, 40%, 31%, and 24% in the long?term future, indicating that available areas to plant Korean pine will be reduced in the future. In conclusion, climate change scenarios and species distribution models (SDMs) create uncertainties in the evaluation of the future distribution of the Korean pine. Therefore, when estimating species distribution under climate change, uncertainties should be considered. In addition, the models show that the suitable habitat area for the Korean pine will decrease in the future, making it important for the climate change adaptation plan to reduce this impact. This study is significant in that it considered uncertainties in the SDMs and RCPs scenarios. The results of this study could be important considerations in the process of plantation planning.

Exploring Influence of Role Models on Basketball Participation among Female Adolescents in Korea

YASIN 서울대학교 대학원 2024 국내석사

Exploring Influence of Role Models on Basketball Participation among female adolescents in Korea Md Fahad Yasin Global Sport Management, Department of Physical Education The Graduate School of Education Seoul National University Children's physical, psychological, and social development is greatly influenced by sport. Sport has always been seen by society as a male- dominated field for a number of reasons, including role structure. Due to this, there are disparities for female athletes at all levels, particularly for young girls. Compared to male athletes, they are seen as having far less sport competence. Young female athletes have chosen to stop competing rather than deal with the obstacles society places in their way of doing sports. Young athletes who look up to role models can have tremendous positive effects from their presence. In an effort to succeed and resemble their role model, athletes learn to emulate the traits and behaviors of their role models while learning to overcome any challenges they may encounter. This study explored how the presence of athlete role models affected young female Korean adolescents in terms of sport participation in particular basketball as numerous studies found that role models can increase perceived sport competence which ultimately increased participants' enthusiasm to play sports in the future. The current research used social cognitive theory, self-efficacy theory and social context framework to examine the benefits and intricacies of the modeling relationship between female adolescents and athlete role models. To accomplish this task, the formative experiences of 6 adolescent female who played basketball were examined. Each athlete was interviewed, with each semi-structured interview focusing on how athlete role models have influenced that the participants in their sport participation. The data from these interviews were qualitatively analyzed using phenomenological study. From phenomenological study, a template emerges in which the characteristics the adolescents take from their athlete role models can be seen as an essential construct and which can have lasting effects on their personal development and sport participation. 본 연구에서는 스포츠 세계에서 역할 구조를 포함한 여러 가지 이유로 사회에서 항상 남성이 지배하는 분야로 간주되고 있다는 연구문제에서 시작되었다. 이로 인해 모든 수준의 여성 운동선수, 특히 어린 청소년들에게 기회의 차이가 크며, 여성 청소년 운동선수들은 사회가 스포츠를 하는 데 방해가 되는 장애물을 처리하기보다는 경쟁을 중단하기로 결정하기도 한다. 롤모델이 지각된 스포츠 역량을 증가시킬 수 있고 궁극적으로 향후 스포츠 참여도를 높일 수 있다는 가설에 따라 농구 종목에서 운동선수 롤모델의 존재가 스포츠 참여 측면에서 어떻게 영향을 미치는지 탐구하고자 하였다. 본 연구는 사회 인지 이론, 자기 효능 이론 및 사회적 맥락 틀을 사용하여 여성 청소년과 운동 선수 롤모델 간의 긍정적 관계성과 복잡성을 조사하였다. 이를 위해 농구에 참여하고 있는 여성 청소년 6명을 심층면담 하였다. 반구조화된 인터뷰를 통해 운동선수 롤모델이 참여자의 스포츠 참여에 어떤 영향을 미쳤는지에 초점을 맞춰 자료가 수집되었다. 수집된 자료는 현상학적 연구를 통해 질적으로 분석되었다. 청소년이 운동선수 롤모델을 갖게 되는 것이 필수 구성 요소로 볼 수 있고 개인 발달과 스포츠 참여에 지속적인 영향을 미친다는 의미 있는 자료를 발견할 수 있었다. 롤 모델을 가지고 있는 여성 청소년들은 그들의 존재로 인해 긍정적인 효과를 얻고 있다는 것이었으며, 자신의 분야에서 성공하고 있는 자신의 롤모델을 닮기 위한 노력의 일환으로 여성 청소년들은 자신이 직면할 수 있는 모든 어려움을 극복하는 방법을 배우면서 롤모델의 특성과 행동을 모방하는 방법을 배우고 있었다. 더 많은 여성 선수들을 양성하고 여학생의 스포츠 참여도를 높이기 위해 여성 청소년들이 존경할 만한 유능한 롤모델을 가질 수 있도록 여성 운동선수들의 스포츠 활동을 여러 방법의 매체를 통해 대중화해야 할 것이다.

Attribute Representation in Neural Language Models

Yu, Dian University of California, Davis ProQuest Dissertat 2022 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Neural models, including neural language models and encoder-decoder models, are the backbone of current natural language processing (NLP) research. Large pre-trained models have greatly improved the performance of both language understanding and generation in many NLP tasks. However, information encoded from the pre-trained models cannot translate to target space easily, which typically requires fine-tuning on domain-specific tasks. Due to domain shift between the pre-training data and the task data, it is still challenging for models to adapt to downstream tasks, especially when the training examples are limited such as in few-shot and zero-shot settings. More importantly, the fine-tuned models can only work well for a small number of domains because of diverging from the original pre-trained model, thus are prone to have over-fitting problems with inductive bias. Although scaled up models with billions or trillions of parameters have shown promising performance with prompts and examples, the challenges still remain.In this thesis, we study if we can learn and inject attribute representation to pre-trained neural models to solve the challenges. Different from a black-box model where the parameters contain vast but encrypted world knowledge, the learned attribute representation can guide the model to learn information relevant to the target task, or serve as supplementary information aside from the original parameters. Attributes can be as high level as language representation in a multilingual transfer learning setting, or as low level such as span or ontology representations. This direction is appealing since we can introduce new attributes to pre-trained models without requiring any changes to the original trained model parameters. As we will show, learning attribute representation is efficient in training with both computation and data requirements. Moreover, it is easy to do transfer learning with even only few examples, while maintaining the original model quality. We believe that training attribute representation is a critical step to reduce the gap between neural model pre-training and applying to target tasks.Specifically, we first introduce methods to represent high-level attributes. Those learned attributes can differentiate from other similar attributes so that they can be utilized to transfer useful knowledge across domains and further to control a neural model towards certain understanding and generation directions. Then, we discuss how to represent low-level attributes from pre-trained models. Those attributes can be hidden with pre-trained models and presented by latent representation. The representation can either be used directly for target tasks by identifying significant features, or be incorporated for further model training. Next, apart from more concrete attributes, we propose methods to integrate task specifications for efficient modeling. Those task-specific attributes model the target task directly, bridging model representation and prediction goals precisely, and enabling performance close to or even above human capacity. Lastly, we apply these attribute representations to dialog systems as a case study. We demonstrate how we can represent different aspects of attributes to build a dialog system from scratch smoothly. We present solutions to the most critical challenges in neural language models in general.

Improving Fine-tuning of Language Models with an Emphasis on Isotropy and Rank

정은아 서울대학교 대학원 2024 국내박사

In the field of natural language processing, a strategy to fine-tune pre-trained language models for downstream tasks is a fundamental approach. Among many fine-tuning tasks, learning text embedding (representation) that captures the underlying semantic information of a given text is an essential task. Given the remarkable progress in the linguistic comprehension capabilities of large-scale pre-trained language models (PLMs), there has been a significant surge in the development of text embedding models leveraging these PLMs in recent times. This dissertation focuses on representations of language models such as Bidirectional Encoder Representations from Transformers (BERT) and studies the techniques to improve the performance of text embeddings. We delve into the two text embedding tasks, Dense Retrieval (DR) and sentence embedding. In the realm of information retrieval, DR models encode queries and documents, thereby generating representations for queries and documents. Using these representations, the relevance between the query and the document is determined. However, representations of PLMs are known to follow an anisotropic distribution, which can be undesirable for relevance estimation. We reveal that representations of popular BERT-based DR models such as ColBERT and RepBERT follow an anisotropic distribution. To cope with the problem, we adopt unsupervised post-processing methods of Normalizing Flow and whitening, which can effectively enhance the isotropy of representations, thereby improving the performance of DR models. Furthermore, with post-processing methods, we can significantly improve the performance of DR models for the out-of-distribution tasks where the distribution of the test dataset differs from that of the training dataset. The next task we focus on is the sentence embedding task. Sentence embedding models estimate the semantic similarity between two given sentences by measuring the similarity between two sentences’ representations. Unsupervised learning of sentence embedding aims to learn representations that capture the underlying semantic information of sentences without the need for human annotation. Among numerous unsupervised models for the sentence embedding task, SimCSE has made a significant progress through self-supervised contrastive learning and has become a foundational baseline for subsequent studies in the field. In pursuit of improving sentence embedding performance through self-supervised learning (SSL), we focus on the representations of SimCSE. Through an in-depth exploration of SimCSE's training dynamics, we uncover a strong correlation between representation rank and performance. Building upon this insight, we introduce the Rank Reduction (RR) regularizer to the fine-tuning of SimCSE. Our experiments reveal that RR not only boosts the performance of SimCSE in sentence embedding tasks but also contributes to the model's stability against changes in random seeds. This result offers valuable insights into the relationship between representation rank and SSL performance in natural language processing, potentially benefiting a wide range of applications.

Implicit Models: Theories and Applications

Gu, Fangda University of California, Berkeley ProQuest Disser 2021 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Deep implicit models are very recent developments on deep learning. Traditionally, deep learning methods rely on explicit forward feeding structures. Super deep structures are proposed to give better performance in various domains. Such approaches have posed difficulty in theoretical analysis and under perform shallower models in some domains. By introducing a recursive structure involving solution to an equilibrium equation in the forward feeding, implicit deep models capture the idea of infinitely deep neural networks while preserving simplicity in model representation, allowing theoretical analysis and better connection to previous efforts in math and control communities. Recent works on implicit models have demonstrated state-of-the-art empirical performances. Despite great ambition, implicit models are very new and see theoretical and empirical challenges. From the theoretical aspect, efficient and effective training and evaluation of implicit models are still open problems. The naive training methods for implicit models are highly inefficient. Trivial initialization easily violates the validity conditions for implicit models. Robustness of implicit models are not well studied. From the empirical aspect, there are still a limited number of works on applying implicit models to solve real world problems. Such works also have not demonstrated significant performance boost over deep learning in general. Applications of implicit models in most areas are mostly unexplored even though implicit models fit in the deep learning framework easily. In the dissertation, we introduce our theoretical and empirical contributions on deep implicit models. The presentation of the dissertation is split into two parts. The first part focuses on theoretical foundations for deep implicit models where we research the evaluation, training, and other topics for implicit deep learning and deep learning in general. The second part then explores the applications of deep implicit models and corresponding theories on real world machine learning applications. We show that implicit models can out-perform existing deep learning techniques in a set of tasks, thanks to the implicit structure which resembles infinitely deep neural networks.