京都大学 / 新制・課程博士 / 博士(情報学) / 甲第25420号 / 情博第858号 / 新制||情||144(附属図書館) / 京都大学大学院情報学研究科知能情報学専攻 / (主査)特定教授 黒橋 禎夫, 教授 河原 達也, 教授 鹿島 久嗣 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

The work herein is, in part, motivated by the idea of creating optimized, actuating structures using additive manufacturing processes (AM). By developing a consistent, repeatable method for designing and manufacturing multi-material compliant mechanisms, significant performance improvements can be seen in application, such as increased mechanism deflection. There are three distinct categories of research that contribute to this overall motivating idea: 1) investigation of an appropriate multi-material topology optimization process for multi-material jetting, 2) understanding the role that manufacturing constraints play in the fabrication of complex, optimized structures, and 3) investigation of an appropriate process for embedding actuating elements within material jetted parts. PolyJet material jetting is the focus of this dissertation research as it is one of the only AM processes capable of utilizing multiple material phases (e.g., stiff and flexible) within a single build, making it uniquely qualified for manufacturing complex, multi-material compliant mechanisms. However, there are two limitations with the PolyJet process within this context: 1) there is currently a dearth of understanding regarding both single and multi-material manufacturing constraints in the PolyJet process and 2) there is no robust embedding methodology for the in-situ embedding of foreign actuating elements within the PolyJet process. These two gaps (and how they relate to the field of compliant mechanism design) will be discussed in detail in this dissertation. Specific manufacturing constraints investigated include 1) "design for embedding" considerations, 2) removal of support material from printed parts, 3) self-supporting angle of surfaces, 4) post-process survivability of fine features, 5) minimum manufacturable feature size, and 6) material properties of digital materials with relation to feature size. The key manufacturing process and geometric design factors that influence each of these constraints are experimentally determined, as well as the quantitative limitations that each constraint imposes on design. / Ph. D.

Large language models (LLMs) are very large deep learning models which are retrained on a huge amount of data. Among the LLMs are sentence bidirectional encoder representations from transformers (SBERT) where advanced training methods such as transformer-based denoising autoEncoder (TSDAE), generative query network (GenQ) and an adaption of generative pseudo labelling (GPL) can be applied. This thesis project aims to develop a semantic search tool for Swedish legal judgments in order to overcome the limitations of traditional keyword searches in legal document retrieval. For this aim, a model adept at understanding the semantic nuances of legal language has been developed by leveraging natural language processing (NLP) and fine- tuning LLMs like SBERT, using advanced training methods such as TSDAE, GenQ, and an adaption of GPL. To generate labeled data out of unlabelled data, a GPT3.5 model was used after it was fine-tuned. The generation of labeled data with the use of a generative model was crucial for this project to train the SBERT efficiently. The search tool has been evaluated. The evaluation demonstrates that the search tool can accurately retrieve relevant documents based on semantic queries and simnifically improve the efficiency and accuracy of legal research. GenQ has been shown to be the most efficient training method for this use case.

The analysis of signals stemming from a physical system is crucial for the experimental investigation of the underlying dynamics that drives the system itself. The field of time series analysis comprises a wide variety of techniques developed with the purpose of characterizing signals and, ultimately, of providing insights on the phenomena that govern the temporal evolution of the generating system. A renowned example in this field is given by spectral analysis: the use of Fourier or Laplace transforms to bring time-domain signals into the more convenient frequency space allows to disclose the key features of linear systems. A more complex scenario turns up when nonlinearity intervenes within a system's dynamics. Nonlinear coupling between a system's degrees of freedom brings about interesting dynamical regimes, such as self-sustained periodic (though anharmonic) oscillations ("limit cycles"), or quasi-periodic evolutions that exhibit sharp spectral lines while lacking strict periodicity ("limit tori"). Among the consequences of nonlinearity, the onset of chaos is definitely the most fascinating one. Chaos is a dynamical regime characterized by unpredictability and lack of periodicity, despite being generated by deterministic laws. Signals generated by chaotic dynamical systems appear as irregular: the corresponding spectra are broad and flat, prediction of future values is challenging, and evolutions within the systems' state spaces converge to strange attractor sets with noninteger dimensionality. Because of these properties, chaotic signals can be mistakenly classified as noise if linear techniques such as spectral analysis are used. The identification of chaos and its characterization require the assessment of dynamical invariants that quantify the complex features of a chaotic system's evolution. For example, Lyapunov exponents provide a marker of unpredictability; the estimation of attractor dimensions, on the other hand, highlights the unconventional geometry of a chaotic system's state space. Nonlinear time series analysis techniques act directly within the state space of the system under investigation. However, experimentally, full access to a system's state space is not always available. Often, only a scalar signal stemming from the dynamical system can be recorded, thus providing, upon sampling, a scalar sequence. Nevertheless, by virtue of a fundamental theorem by Takens, it is possible to reconstruct a proxy of the original state space evolution out of a single, scalar sequence. This reconstruction is carried out by means of the so-called embedding procedure: m-dimensional vectors are built by picking successive elements of the scalar sequence delayed by a lag L. On the other hand, besides posing some necessary conditions on the integer embedding parameters m and L, Takens' theorem does not provide any clue on how to choose them correctly. Although many optimal embedding criteria were proposed, a general answer to the problem is still lacking. As a matter of fact, conventional methods for optimal embedding are flawed by several drawbacks, the most relevant being the need for a subjective evaluation of the outcomes of applied algorithms. Tackling the issue of optimally selecting embedding parameters makes up the core topic of this thesis work. In particular, I will discuss a novel approach that was pursued by our research group and that led to the development of a new method for the identification of suitable embedding parameters. Rather than most conventional approaches, which seek a single optimal value for m and L to embed an input sequence, our approach provides a set of embedding choices that are equivalently suitable to reconstruct the dynamics. The suitability of each embedding choice m, L is assessed by relying on statistical testing, thus providing a criterion that does not require a subjective evaluation of outcomes. The starting point of our method are embedding-dependent correlation integrals, i.e. cumulative distributions of embedding vector distances, built out of an input scalar sequence. In the case of Gaussian white noise, an analytical expression for correlation integrals is available, and, by exploiting this expression, a gauge transformation of distances is introduced to provide a more convenient representation of correlation integrals. Under this new gauge, it is possible to test—in a computationally undemanding way—whether an input sequence is compatible with Gaussian white noise and, subsequently, whether the sequence is compatible with the hypothesis of an underlying chaotic system. These two statistical tests allow ruling out embedding choices that are unsuitable to reconstruct the dynamics. The estimation of correlation dimension, carried out by means of a newly devised estimator, makes up the third stage of the method: sets of embedding choices that provide uniform estimates of this dynamical invariant are deemed to be suitable to embed the sequence.The method was successfully applied to synthetic and experimental sequences, providing new insight into the longstanding issue of optimal embedding. For example, the relevance of the embedding window (m-1)L, i.e. the time span covered by each embedding vector, is naturally highlighted by our approach. In addition, our method provides some information on the adequacy of the sampling period used to record the input sequence.The method correctly distinguishes a chaotic sequence from surrogate ones generated out of it and having the same power spectrum. The technique of surrogate generation, which I also addressed during my Ph. D. work to develop new dedicated algorithms and to analyze brain signals, allows to estimate significance levels in situations where standard analytical algorithms are unapplicable. The novel embedding approach being able to tell apart an original sequence from surrogate ones shows its capability to distinguish signals beyond their spectral—or autocorrelation—similarities.One of the possible applications of the new approach concerns another longstanding issue, namely that of distinguishing noise from chaos. To this purpose, complementary information is provided by analyzing the asymptotic (long-time) behaviour of the so-called time-dependent divergence exponent. This embedding-dependent metric is commonly used to estimate—by processing its short-time linearly growing region—the maximum Lyapunov exponent out of a scalar sequence. However, insights on the kind of source generating the sequence can be extracted from the—usually overlooked—asymptotic behaviour of the divergence exponent. Moreover, in the case of chaotic sources, this analysis also provides a precise estimate of the system's correlation dimension. Besides describing the results concerning the discrimination of chaotic systems from noise sources, I will also discuss the possibility of using the related correlation dimension estimates to improve the third stage of the method introduced above for the identification of suitable embedding parameters. The discovery of chaos as a possible dynamical regime for nonlinear systems led to the search of chaotic behaviour in experimental recordings. In some fields, this search gave plenty of positive results: for example, chaotic dynamics was successfully identified and tamed in electronic circuits and laser-based optical setups. These two families of experimental chaotic systems eventually became versatile tools to study chaos and its possible applications. On the other hand, chaotic behaviour is also looked for in climate science, biology, neuroscience, and even economics. In these fields, nonlinearity is widespread: many smaller units interact nonlinearly, yielding a collective motion that can be described by means of few, nonlinearly coupled effective degrees of freedom. The corresponding recorded signals exhibit, in many cases, an irregular and complex evolution. A possible underlying chaotic evolution—as opposed to a stochastic one—would be of interest both to reveal the presence of determinism and to predict the system's future states. While some claims concerning the existence of chaos in these fields have been made, most results are debated or inconclusive. Nonstationarity, low signal-to-noise ratio, external perturbations and poor reproducibility are just few among the issues that hinder the search of chaos in natural systems. In the final part of this work, I will briefly discuss the problem of chasing chaos in experimental recordings by considering two example sequences, the first one generated by an electronic circuit and the second one corresponding to recordings of brain activity. The present thesis is organized as follows. The core concepts of time series analysis, including the key features of chaotic dynamics, are presented in Chapter 1. A brief review of the search for chaos in experimental systems is also provided; the difficulties concerning this quest in some research fields are also highlighted. Chapter 2 describes the embedding procedure and the issue of optimally choosing the related parameters. Thereupon, existing methods to carry out the embedding choice are reviewed and their limitations are pointed out. In addition, two embedding-dependent nonlinear techniques that are ordinarily used to characterize chaos, namely the estimation of correlation dimension by means of correlation integrals and the assessment of maximum Lyapunov exponent, are presented. The new approach for the identification of suitable embedding parameters, which makes up the core topic of the present thesis work, is the subject of Chapter 3 and 4. While Chapter 3 contains the theoretical outline of the approach, as well as its implementation details, Chapter 4 discusses the application of the approach to benchmark synthetic and experimental sequences, thus illustrating its perks and its limitations. The study of the asymptotic behaviour of the time-dependent divergent exponent is presented in Chapter 5. The alternative estimator of correlation dimension, which relies on this asymptotic metric, is discussed as a possible improvement to the approach described in Chapters 3, 4. The search for chaos out of experimental data is discussed in Chapter 6 by means of two examples of real-world recordings. Concluding remarks are finally drawn in Chapter 7.

Cette thèse porte sur les processus croissants pour l'ordre convexe que nous désignons sous le nom de peacocks. Un résultat remarquable dû à Kellerer stipule qu'un processus stochastique à valeurs réelles est un peacock si et seulement s'il possède les mêmes marginales unidimensionnelles qu'une martingale. Une telle martingale est dite associée à ce processus. Mais dans son article, Kellerer ne donne ni d'exemple de peacock, ni d'idée précise sur la construction d'une martingale associée pour un peacock donné. Ainsi, comme d'autres travaux sur les peacocks, notre étude vise deux objectifs. Il s'agit d'exhiber de nouvelles familles de peacocks et de construire des martingales associées pour certains peacocks. Dans les trois premiers chapitres, nous exhibons diverses classes de peacocks en utilisant successivement les notions de monotonie conditionnelle, de peacock très fort et de positivité totale d'ordre 2. En particulier, nous fournissons plusieurs extensions du résultat de Carr-Ewald-Xiao selon lequel la moyenne arithmétique du mouvement brownien géométrique, encore appelée "option asiatique" est un peacock. L'objet du dernier chapitre est de construire des martingales associées pour une classe de peacocks. Pour cela, nous utilisons les plongements d'Azéma-Yor et de Bertoin-Le Jan. L'originalité de ce chapitre est l'utilisation de la positivité totale d'ordre 2 dans l'étude du plongement d'Azéma-Yor / This thesis deals with real valued stochastic processes which increase in the convex order. We call them peacocks. A remarkable result due to Kellerer states that a real valued process is a peacock if and only if it has the same one-dimensional marginals as a martingale. Such a martingale is said to be associated to this process. But in his article, Kellerer provides neither an example of peacock nor a concrete idea to construct an associated martingale to a given peacock. Hence, as other investigations on peacocks, our study has two purposes. We first exhibit new families of peacocks and then, we contruct associated martingales to certain of them. In the first three chapters, we exhibit several classes of peacocks using successively the notions of conditional monotonicity, very strong peacock and total positivity of order 2. In particular, we provide many extensions of Carr-Ewald-Xiao result which states that the arithmetic mean of geometric Brownian motion, also called "Asian option" is a peacock. The purpose of the last chapter is to construct associated martingales to certain peacocks. To this end, we use Azéma-Yor and Bertoin-Le Jan embedding algorithms. The originality of this chapter is the use of total positivity of order 2 in the study of Azéma-Yor embedding algorithm

In this thesis we study two different topics, both in the context of semigroup constructions. The first is the investigation of an embedding problem, specifically the problem of whether it is possible to embed any given finitely presentable semigroup into a D-simple finitely presentable semigroup. We consider some well-known semigroup constructions, investigating their properties to determine whether they might prove useful for finding a solution to our problem. We carry out a more detailed study into a more complicated semigroup construction, the Byleen extension, which has been used to solve several other embedding problems. We prove several results regarding the structure of this extension, finding necessary and sufficient conditions for an extension to be D-simple and a very strong necessary condition for an extension to be finitely presentable. The second topic covered in this thesis is relative rank, specifically the sequence obtained by taking the rank of incremental direct powers of a given semigroup modulo the diagonal subsemigroup. We investigate the relative rank sequences of infinite Cartesian products of groups and of semigroups. We characterise all semigroups for which the relative rank sequence of an infinite Cartesian product is finite, and show that if the sequence is finite then it is bounded above by a logarithmic function. We will find sufficient conditions for the relative rank sequence of an infinite Cartesian product to be logarithmic, and sufficient conditions for it to be constant. Chapter 4 ends with the introduction of a new topic, relative presentability, which follows naturally from the topic of relative rank.

Dans cette thèse, nous nous intéressons à plusieurs problèmes liés à la conception de réseaux géométriques et aux plongements isométriques dans le plan.Nous commençons par étudier la généralisation du problème du réseau de Manhattan classique aux plans normés. Étant donné un ensemble de terminaux, nous recherchons le réseau de longueur totale minimum qui connecte chaque paire de terminaux par un plus court chemin dans la métrique définie par la norme. Nous proposons un algorithme d'approximation facteur 2.5 pour ce problème en temps O(mn^3) avec n le nombre de terminaux et m le nombre de directions de la boule unitaire. Le deuxième problème étudié est une version orientée des réseaux de Manhattan dont le but est de construire un réseau orienté de taille minimum dans lequel pour chaque paire de terminaux u, v est relié par un plus court chemin rectilinéaire de u vers v et un autre de v vers u. Nous proposons un algorithme d'approximation facteur 2 pour ce problème en temps O(n^3) où n est le nombre de terminaux.Nous nous intéressons ensuite à la recherche d'un spanner (un sous-graphe approximant les distances) planaire pour les graphes de disques unitaires (UDG) qui modélise les réseaux ad hoc sans fils. Nous présentons un algorithme qui construit un spanner planaire avec un facteur d'étirement constant en terme de distance de graphe pour UDG. Cet algorithme utilise uniquement des propriétés locales et peut donc être implémenté de manière distribuée.Finalement nous étudions le problème de la reconnaissance des espaces plongeables isométriquement dans le plan l_1 pour lequel nous proposons un algorithme en temps optimal O(n^2) pour sa résolution, ainsi que la généralisation de ce problème aux plans normés dont la boule unitaire est un polygone convexe central symétrique. / In this thesis, we study several problems related to the design of geometric networks and isometric embeddings into the plane.We start by considering the generalization of the classical Minimum Manhattan Network problem to all normed planes. We search the minimum network that connects each pair of terminals by a shortest path in this norm. We propose a factor 2.5 approximation algorithm in time O(mn^3), where n is the number of terminals and m is the number of directions of the unit ball.The second problem presented is an oriented version of the minumum Manhattan Network problem, we want to obtain a minimum oriented network such that for each pair u, v of terminals, there is a shortest rectilinear path from u to v and another path from v to u.We describe a factor 2 approximation algorithm with complexity O(n^3) where n is the number of terminals for this problem.Then we study the problem of finding a planar spanner (a subgraph which approximates the distances) of the Unit Disk Graph (UDG) which is used to modelize wireless ad hoc networks. We present an algorithm for computing a constant hop stretch factor planar spanner for all UDG. This algorithm uses only local properties and it can be implemented in distributed manner.Finally, we study the problem of recognizing metric spaces that can be isometrically embbed into the rectilinear plane and we provide an optimal time O(n^2) algorithm to solve this problem. We also study the generalization of this problem to all normed planes whose unit ball is a centrally symmetric convex polygon.

MASTER THESIS Petr Lukeš Emergence of space geometries from quantum entanglement Institute of Theoretical Physics Supervisor of the master thesis: Mgr. Martin Scholtz, Ph.D. Study programme: Physics Study branch: Theoretical physics Prague 2019 Abstract: Connecting the field of Quantum Physics and General Relativity is one of the main interests of contemporary Theoretical Physics. This work attempts to find solution to simplified version of this problem. Firstly entropy is shown to be a good meeting point between the two different theories. Then some of entropy's less intuitive properties are shown, namely its dependence on area, not volume. This relation is studied from both Relativistic and Quantum viewpoint. After- wards there is a short description of a quantum model interpretable as geometry based on the information between its subsystems. Lastly, results of computations within this model are presented.

Le graphe est un concept puissant pour la représentation des relations entre des paires d'entités. Les données ayant une structure de graphes sous-jacente peuvent être trouvées dans de nombreuses disciplines, décrivant des composés chimiques, des surfaces des modèles tridimensionnels, des interactions sociales ou des bases de connaissance, pour n'en nommer que quelques-unes. L'apprentissage profond (DL) a accompli des avancées significatives dans une variété de tâches d'apprentissage automatique au cours des dernières années, particulièrement lorsque les données sont structurées sur une grille, comme dans la compréhension du texte, de la parole ou des images. Cependant, étonnamment peu de choses ont été faites pour explorer l'applicabilité de DL directement sur des données structurées sous forme des graphes. L'objectif de cette thèse est d'étudier des architectures de DL sur des graphes et de rechercher comment transférer, adapter ou généraliser à ce domaine des concepts qui fonctionnent bien sur des données séquentielles et des images. Nous nous concentrons sur deux primitives importantes : le plongement de graphes ou leurs nœuds dans une représentation de l'espace vectorielle continue (codage) et, inversement, la génération des graphes à partir de ces vecteurs (décodage). Nous faisons les contributions suivantes. Tout d'abord, nous introduisons Edge-Conditioned Convolutions (ECC), une opération de type convolution sur les graphes réalisés dans le domaine spatial où les filtres sont générés dynamiquement en fonction des attributs des arêtes. La méthode est utilisée pour coder des graphes avec une structure arbitraire et variable. Deuxièmement, nous proposons SuperPoint Graph, une représentation intermédiaire de nuages de points avec de riches attributs des arêtes codant la relation contextuelle entre des parties des objets. Sur la base de cette représentation, l'ECC est utilisé pour segmenter les nuages de points à grande échelle sans sacrifier les détails les plus fins. Troisièmement, nous présentons GraphVAE, un générateur de graphes permettant de décoder des graphes avec un nombre de nœuds variable mais limité en haut, en utilisant la correspondance approximative des graphes pour aligner les prédictions d'un auto-encodeur avec ses entrées. La méthode est appliquée à génération de molécules / Graph is a powerful concept for representation of relations between pairs of entities. Data with underlying graph structure can be found across many disciplines, describing chemical compounds, surfaces of three-dimensional models, social interactions, or knowledge bases, to name only a few. There is a natural desire for understanding such data better. Deep learning (DL) has achieved significant breakthroughs in a variety of machine learning tasks in recent years, especially where data is structured on a grid, such as in text, speech, or image understanding. However, surprisingly little has been done to explore the applicability of DL on graph-structured data directly.The goal of this thesis is to investigate architectures for DL on graphs and study how to transfer, adapt or generalize concepts working well on sequential and image data to this domain. We concentrate on two important primitives: embedding graphs or their nodes into a continuous vector space representation (encoding) and, conversely, generating graphs from such vectors back (decoding). To that end, we make the following contributions.First, we introduce Edge-Conditioned Convolutions (ECC), a convolution-like operation on graphs performed in the spatial domain where filters are dynamically generated based on edge attributes. The method is used to encode graphs with arbitrary and varying structure.Second, we propose SuperPoint Graph, an intermediate point cloud representation with rich edge attributes encoding the contextual relationship between object parts. Based on this representation, ECC is employed to segment large-scale point clouds without major sacrifice in fine details.Third, we present GraphVAE, a graph generator allowing to decode graphs with variable but upper-bounded number of nodes making use of approximate graph matching for aligning the predictions of an autoencoder with its inputs. The method is applied to the task of molecule generation

James W. Alexander, no artigo[1],mostra que se tivermos um mergulho PL f : \'S POT. 1\' × \'S POT. 1\' \'S POT. 3\', então o fecho de uma das componentes conexas de \'S POT. 3\' f(\'S POT. 1\' × \'S POT. 1\') é homeomorfo a um toro sólido, isto é, homeomorfo a \'S POT. 1\' × \'D POT. 2\'. Este teorema ficou conhecido por Teorema do toro de Alexander. Nesta dissertação, estamos detalhando a demonstração deste teorema feita em[25] que é diferente da demonstração apresentada em [1]. Mais geralmente, para um mergulho diferenciável f : \'S POT. p\' × \'S POT. q\' \'S POT. p + q+1\' , demonstra-se que o fecho de uma das componentes conexasde \'S POT. p +q + 1\' f(\'S POT. p\' × \'S POT. q\') é difeomorfo a \'S POT. p\' × \'D POT. q + 1\' se p q 1 e p + q \'DIFERENTE DE\' 3 ou se p = 2 e q = 1 um dos fechos será homeomorfo a \'S POT. 2\' × \'D POT. 2\' , nesta dissertação estaremos também detalhando estas demonstrações feita em [20] / James W. Alexander shows in[1] that the closure of one of the two connected components of \'S POT. 3\'f( \'S POT. 1 × \'S POT. 1\') is homeomorphic to a solid torus \'S POT. 1\' × \'D POT. 2\' , where f : \'S POT. 1\' ×\' SPOT. 1\' \'S POT. 3\' is a PL embedding. This result became known as Alexanders torus theorem. In this dissertation we are detailing the proof of this theorem made in[25] which is different from the demonstration presented in[1]. More generally, when considering a smooth embeding f : \'S POT. p\' × \'S POT. q\' \' SPOT. p+q+1\' , it is demonstrated that the closure of one of the two connected components \'S POT. p+q+1\' f (\'S POT. p\' × \'S POT. q\' ) is diffeomorphic to \'S POT. p\' × \'D POT. q+1\' if p q 1 and p+q \'DIFFERENT OF\' 3 or if p = 2 and q = 1 one of the closures will be homeomorphic to \'S POT. 2\' × \'D POT. 2\'. In this work we are also detailing the proves made in[20]