Orientador: Marcelo Takeshi Yamashita / Banca: Tobias Frederico / Banca: Renato Higa / Resumo: Neste trabalho investigamos a universalidade em sistemas de três e quatro bósons através do cálculo das suas energias de ligação e dos raios quadráticos médios. Utilizando duas funções de escala calculadas com um potencial de alcance zero e um potencial de alcance finito corrigimos em primeira ordem em r0/a (r0 e a são, respectivamente, o alcance efetivo do potencial e o comprimento de espalhamento de dois corpos) o ponto onde os estados excitados de três corpos desaparecem. Estudamos também as estruturas dos estados de quatro corpos associados ao estado fundamental de três corpos para energia de dois corpos igual a zero. Esses estados são formados predominantemente por uma configuração do tipo 3+1. Os cálculos foram realizados no espaço das configurações usando um método variacional / Abstract: In this work we investigated the universality in three- and four-boson systems calculating their energies and root-mean-square radii. Using two scaling functions calculated with a zero and a finite range potentials, we corrected to first order in r0/a (r0 and a are, respectively, the effective range of the potential and the two-body scattering length) the point where the three-body excited states disappear. We also studied the structures of the four-body statestied to the three-body ground state for a two-body energy equal zero. These states are predominantly composed by a 3+1 configuration. The calculations were performed in the configuration space using a variational method / Mestre

Orientador: Marcelo Takeshi Yamashita / Banca: Lauro Tomio / Banca: Marijana Brtka / Resumo: Este trabalho foca no estudo de sistemas de poucos corpos em duas dimensões no regime universal, onde as propriedades do sistema quântico independem dos detalhes da interação de curto alcance entre as partículas (o comprimento de espalhamento de dois corpos é muito maior que o alcance do potencial). Nós utilizamos a decomposição de Faddeev para escrever as equações para os estados ligados. Através da solução numérica dessas equações nós calculamos as energias de ligação e os raios quadráticos médios de um sistema composto por dois bósons (A) e uma partícula diferente (B). Para uma razão de massas mB/mA = 0.01 o sistema apresenta oito estados ligados de três corpos, os quais desaparecem um por um conforme aumentamos a razão de massas restando somente os estados fundamental e primeiro excitado. Os comportamentos das energias e dos raios para razões de massa pequenas podem ser entendidos através de um potencial do tipo Coulomb a curtas distâncias (onde o estado fundamental está localizado) que aparece quando utilizamos uma aproximação de Born-Oppenheimer. Para grandes razões de massa os dois estados ligados restantes são consistentes com uma estrutura de três corpos mais simétrica. Nós encontramos que no limiar da razão de massas em que os estados desaparecem os raios divergem linearmente com as energias de três corpos escritas em relação ao limiar de dois corpos / Abstract: This work is focused in the study of two dimensional few-body physics in the universal regime, where the properties of the quantum system are independent on the details of the short-range interaction between particles (the two-body scatter- ing length is much larger than the range of the potential). We used the Faddeev decomposition to write the bound-state equations and we calculated the three-body binding energies and root-mean-square (rms) radii for a three-body system in two dimensions compounded by two identical bosons (A) and a different particle (B). For mass ratio mB/mA = 0.01 the system displays eight three-body bound states, which disappear one by one as the mass ratio is increased leaving only the ground and the first excited states. Energies and radii of the states for small mass ratios can be understood quantitatively through the Coulomb-like Born-Oppenheimer potential at small distances where the lowest-lying of these states are located. For large mass ratio the radii of the two remaining bound states are consistent with a more sym- metric three-body structure. We found that the radii diverge linearly at the mass ratio threshold where the three-body excited states disappear. The divergences are linear in the inverse energy deviations from the corresponding two-body thresholds / Mestre

The analysis of scattering data is usually done by fitting the S-matrix at real experimental energies. An analytic continuation to complex and negative energies must then be performed to locate possible resonances and bound states, which correspond to poles of the S-matrix. Difficulties in the analytic continuation arise since the S-matrix is energy dependent via the momentum, k and the Sommerfeld parameter, η, which makes it multi-valued. In order to circumvent these difficulties, in this work, the S-matrix is written in a semi-analytic form in terms of the Jost matrices, which can be given as a product of known functions dependent on k and η, and unknown functions that are entire and singled-valued in energy. The unknown functions are approximated by truncated Taylor series where the expansion coefficients serve as the data-fitting parameters. The proper analytic structure of the S-matrix is thus maintained. This method is successfully tested with data generated by a model scattering potential. It is then applied to α12C scattering, where resonances of 16O in the quantum states Jρ =0+, 1−, 2+, 3−, and 4+ are located. The parameters of these resonances are accurately determined, as well as the corresponding S-matrix residues and Asymptotic Normalisation Coefficients, relevant to astrophysics. The method is also applied to dα scattering to determine the bound and resonance state parameters, corresponding S-matrix residues and Asymptotic Normalisation Coefficients of 6Li in the 1+, 2+, 3+, 2−, and 3− states. / Thesis (PhD)--University of Pretoria, 2020. / National Research Foundation (NRF) / Physics / PhD / Unrestricted

Graphs are commonly employed to model complex data and discover latent patterns and relationships between entities in the real world. Canonical graph learning models have achieved remarkable progress in modeling and inference on graph-structured data that consists of nodes connected by edges. Generally, they leverage abundant labeled data for model training and thus inevitably suffer from the label scarcity issue due to the expense and hardship of data annotation in practice. Data-efficient graph learning attempts to address the prevailing data scarcity issue in graph mining problems, of which the key idea is to transfer knowledge from the related resources to obtain the models with good generalizability to the target graph-related tasks with mere annotations. However, the generalization of the models to data-scarce scenarios is faced with challenges including 1) dealing with graph structure and structural heterogeneity to extract transferable knowledge; 2) selecting beneficial and fine-grained knowledge for effective transfer; 3) addressing the divergence across different resources to promote knowledge transfer. Motivated by the aforementioned challenges, the dissertation mainly focuses on three perspectives, i.e., knowledge extraction with graph heterogeneity, knowledge selection, and knowledge transfer. The purposed models are applied to various node classification and graph classification tasks in the low-data regimes, evaluated on a variety of datasets, and have shown their effectiveness compared with the state-of-the-art baselines.

No description available.

No description available.

京都大学 / 新制・課程博士 / 博士(理学) / 甲第23701号 / 理博第4791号 / 新制||理||1686(附属図書館) / 京都大学大学院理学研究科物理学・宇宙物理学専攻 / (主査)教授 永江 知文, 准教授 成木 恵, 教授 中家 剛 / 学位規則第4条第1項該当 / Doctor of Science / Kyoto University / DFAM

Side-polished fiber allows access to the evanescent field propagating in the cladding of a few-mode fiber. This cladding mode is analyzed and experimentally validated to further the design of a novel class of fiber optic devices. To do this, specific modes are excited in the polished fiber using a phase-only spatial light modulator to determine spatial mode distribution. Each mode is excited and compared to the expected field distribution and to confirm that higher order modes can propagate through side-polished fiber. Based on each mode’s distribution, a side-polished fiber can be designed so that perturbations on the polished portion of the fiber effect each mode independently. By carefully analyzing the effects of identical perturbations on each mode, it is determined that each mode can be isolated based on the geometry of the polished fiber and careful alignment of the mode field. This research has the potential to advance the development of novel fiber-based sensors and communications devices utilizing mode-based interferometry and mode multiplexing. / M.S. / Fiber optic devices have seen significant advancement since the realization of the laser and low-loss optical fiber. Modern day fiber optics are commonly utilized for high-bandwidth communications and specialized sensing applications. Utilizing multiple modes, or wave distributions, in a fiber provides significant advantages towards increasing bandwidth for communications and provides potential for more accurate sensing techniques. Significant research has been conducted in both the sensing and communication field, but mode-domain devices have the capability to significantly advance the field of fiber optic devices. This thesis demonstrates the potential for side-polished fiber geometry to effect each mode independently, thus allowing side-polished fiber to be utilized for realizing novel devices such as multiplexing devices and fiber optic sensors.

Long-period fiber gratings (LPGs) with a period ranging from several hundred micrometers to a few millimeters can couple a core mode to discrete co-propagating cladding modes when the phase matching condition is satisfied. The rapid attenuation of cladding modes results in loss bands in the transmission spectrum. As the attenuation bands are sensitive to the LPG period and the fiber surrounding environment such as temperature, strain and ambient refractive index, LPGs can be used for sensing. However, traditional LPGs with gratings inscribed in the fibers can only sense a single point and cannot be used for distributed sensing. Although new ideas were proposed to use traveling LPG formed by a pulsed acoustic wave, the large attenuation of the acoustic wave in the fiber greatly limits the sensing range to only several meters. In this thesis, we proposed to use a traveling LPG formed by the interference of two high power co-propagating core modes, usually LP01 and LP11. The beating of the two modes will induce a refractive index grating due to the optical Kerr effect, and the grating is called optically induced long-period fiber grating (OLPG). Compared to the grating induced by acoustic waves, OLPG is able to travel for a long distance due to the small attenuation of the guided core modes. Mode conversion in the OLPG is numerically simulated and analyzed using the finite-difference beam propagation method (FD-BPM). The result shows full conversion for both core-core and core-cladding mode coupling under phase matching condition. Moreover, the sensitivity of OLPG to temperature, axial strain and ambient refractive index is investigated and analyzed. It is seen that the sensitivities of temperature and axial strain with OLPG are different from the traditional LPGs since the period variation in OLPG is caused by the effective index difference of the two core modes at the writing wavelength, while in the traditional LPGs it is directly induced by temperature or strain. For the refractive index sensitivity with a large cladding, OLPG behaves the same as a traditional LPG with only material contributions since the grating period remains unchanged. / Master of Science

Recent advances in large neural network-style models have demonstrated great performance in various applications, such as image generation, question answering, and audio classification. However, these deep and high-capacity models require a large amount of labeled data to function properly, rendering them inapplicable in many real-world scenarios. This dissertation focuses on the development and evaluation of advanced machine learning algorithms to solve the following research questions: (1) How to learn novel classes with limited labeled data, (2) How to adapt a large pre-trained model to the target domain if only unlabeled data is available, (3) How to boost the performance of the few-shot learning model with unlabeled data, and (4) How to utilize limited labeled data to learn new classes without the training data in the same domain. First, we study few-shot learning in text classification tasks. Meta-learning is becoming a popular approach for addressing few-shot text classification and has achieved state-of-the-art performance. However, the performance of existing approaches heavily depends on the interclass variance of the support set. To address this problem, we propose a TART network for few-shot text classification. The model enhances the generalization by transforming the class prototypes to per-class fixed reference points in task-adaptive metric spaces. In addition, we design a novel discriminative reference regularization to maximize divergence between transformed prototypes in task-adaptive metric spaces to improve performance further. In the second problem we focus on self-learning in cross-lingual transfer task. Our goal here is to develop a framework that can make the pretrained cross-lingual model continue learning the knowledge with large amount of unlabeled data. Existing self-learning methods in crosslingual transfer tasks suffer from the large number of incorrectly pseudo-labeled samples used in the training phase. We first design an uncertainty-aware cross-lingual transfer framework with pseudo-partial-labels. We also propose a novel pseudo-partial-label estimation method that considers prediction confidences and the limitation to the number of candidate classes. Next, to boost the performance of the few-shot learning model with unlabeled data, we propose a semi-supervised approach for few-shot semantic segmentation task. Existing solutions for few-shot semantic segmentation cannot easily be applied to utilize image-level weak annotations. We propose a class-prototype augmentation method to enrich the prototype representation by utilizing a few image-level annotations, achieving superior performance in one-/multi-way and weak annotation settings. We also design a robust strategy with softmasked average pooling to handle the noise in image-level annotations, which considers the prediction uncertainty and employs the task-specific threshold to mask the distraction. Finally, we study the cross-domain few-shot learning in the semantic segmentation task. Most existing few-shot segmentation methods consider a setting where base classes are drawn from the same domain as the new classes. Nevertheless, gathering enough training data for meta-learning is either unattainable or impractical in many applications. We extend few-shot semantic segmentation to a new task, called Cross-Domain Few-Shot Semantic Segmentation (CD-FSS), which aims to generalize the meta-knowledge from domains with sufficient training labels to low-resource domains. Then, we establish a new benchmark for the CD-FSS task and evaluate both representative few-shot segmentation methods and transfer learning based methods on the proposed benchmark. We then propose a novel Pyramid-AnchorTransformation based few-shot segmentation network (PATNet), in which domain-specific features are transformed into domain-agnostic ones for downstream segmentation modules to fast adapt to unseen domains. / Doctor of Philosophy / Nowadays, deep learning techniques play a crucial role in our everyday existence. In addition, they are crucial to the success of many e-commerce and local businesses for enhancing data analytics and decision-making. Notable applications include intelligent transportation, intelligent healthcare, the generation of natural language, and intrusion detection, among others. To achieve reasonable performance on a new task, these deep and high-capacity models require thousands of labeled examples, which increases the data collection effort and computation costs associated with training a model. Moreover, in many disciplines, it might be difficult or even impossible to obtain data due to concerns such as privacy and safety. This dissertation focuses on learning with limited labeled data in natural language processing and computer vision tasks. To recognize novel classes with a few examples in text classification tasks, we develop a deep learning-based model that can capture both cross- task transferable knowledge and task-specific features. We also build an uncertainty-aware self-learning framework and a semi-supervised few-shot learning method, which allow us to boost the pre-trained model with easily accessible unlabeled data. In addition, we propose a cross-domain few-shot semantic segmentation method to generalize the model to different domains with a few examples. By handling these unique challenges in learning with limited labeled data and developing suitable approaches, we hope to improve the efficiency and generalization of deep learning methods in the real world.