Cette thèse aborde le problème du manque de performance des outils exploitant des représentationsà base de graphes en reconnaissance des formes. Nous proposons de contribuer aux nouvellesméthodes proposant de tirer partie, à la fois, de la richesse des méthodes structurelles et de la rapidité des méthodes de reconnaissance de formes statistiques. Deux principales contributions sontprésentées dans ce manuscrit. La première correspond à la proposition d'une nouvelle méthode deprojection explicite de graphes procédant par analyse multi-facettes des graphes. Cette méthodeeffectue une caractérisation des graphes suivant différents niveaux qui correspondent, selon nous,aux point-clés des représentations à base de graphes. Il s'agit de capturer l'information portéepar un graphe au niveau global, au niveau structure et au niveau local ou élémentaire. Ces informationscapturées sont encapsulés dans un vecteur de caractéristiques numériques employantdes histogrammes flous. La méthode proposée utilise, de plus, un mécanisme d'apprentissage nonsupervisée pour adapter automatiquement ses paramètres en fonction de la base de graphes àtraiter sans nécessité de phase d'apprentissage préalable. La deuxième contribution correspondà la mise en place d'une architecture pour l'indexation de masses de graphes afin de permettre,par la suite, la recherche de sous-graphes présents dans cette base. Cette architecture utilise laméthode précédente de projection explicite de graphes appliquée sur toutes les cliques d'ordre 2pouvant être extraites des graphes présents dans la base à indexer afin de pouvoir les classifier.Cette classification permet de constituer l'index qui sert de base à la description des graphes etdonc à leur indexation en ne nécessitant aucune base d'apprentissage pré-étiquetées. La méthodeproposée est applicable à de nombreux domaines, apportant la souplesse d'un système de requêtepar l'exemple et la granularité des techniques d'extraction ciblée (focused retrieval). / This thesis addresses the problem of lack of efficient computational tools for graph based structural pattern recognition approaches and proposes to exploit computational strength of statistical pattern recognition. It has two fold contributions. The first contribution is a new method of explicit graph embedding. The proposed graph embedding method exploits multilevel analysis of graph for extracting graph level information, structural level information and elementary level information from graphs. It embeds this information into a numeric feature vector. The method employs fuzzy overlapping trapezoidal intervals for addressing the noise sensitivity of graph representations and for minimizing the information loss while mapping from continuous graph space to discrete vector space. The method has unsupervised learning abilities and is capable of automatically adapting its parameters to underlying graph dataset. The second contribution is a framework for automatic indexing of graph repositories for graph retrieval and subgraph spotting. This framework exploits explicit graph embedding for representing the cliques of order 2 by numeric feature vectors, together with classification and clustering tools for automatically indexing a graph repository. It does not require a labeled learning set and can be easily deployed to a range of application domains, offering ease of query by example (QBE) and granularity of focused retrieval.

Face à la quantité d'information textuelle disponible sur le web en langue arabe, le développement des Systèmes de Recherche d'Information (SRI) efficaces est devenu incontournable pour retrouver l'information pertinente. La plupart des SRIs actuels de la langue arabe reposent sur la représentation par sac de mots et l'indexation des documents et des requêtes est effectuée souvent par des mots bruts ou des racines. Ce qui conduit à plusieurs problèmes tels que l'ambigüité et la disparité des termes, etc.Dans ce travail de thèse, nous nous sommes intéressés à apporter des solutions aux problèmes d'ambigüité et de disparité des termes pour l'amélioration de la représentation des documents et le processus de l'appariement des documents et des requêtes. Nous apportons quatre contributions au niveau de processus de représentation, d'indexation et de recherche d'information en langue arabe. La première contribution consiste à représenter les documents à la fois par des termes simples et des termes complexes. Cela est justifié par le fait que les termes simples seuls et isolés de leur contexte sont ambigus et moins précis pour représenter le contenu des documents. Ainsi, nous avons proposé une méthode hybride pour l’extraction de termes complexes en langue arabe, en combinant des propriétés linguistiques et des modèles statistiques. Le filtre linguistique repose à la fois sur l'étiquetage morphosyntaxique et la prise en compte des variations pour sélectionner les termes candidats. Pour sectionner les termes candidats pertinents, nous avons introduit une mesure d'association permettant de combiner l'information contextuelle avec les degrés de spécificité et d'unité. La deuxième contribution consiste à explorer et évaluer les systèmes de recherche d’informations permettant de tenir compte de l’ensemble des éléments d’indexation (termes simples et complexes). Par conséquent, nous étudions plusieurs extensions des modèles existants de RI pour l'intégration des termes complexes. En outre, nous explorons une panoplie de modèles de proximité. Pour la prise en compte des dépendances de termes dans les modèles de RI, nous introduisons une condition caractérisant de tels modèle et leur validation théorique. La troisième contribution permet de pallier le problème de disparité des termes en proposant une méthode pour intégrer la similarité entre les termes dans les modèles de RI en s'appuyant sur les représentations distribuées des mots (RDMs). L'idée sous-jacente consiste à permettre aux termes similaires à ceux de la requête de contribuer aux scores des documents. Les extensions des modèles de RI proposées dans le cadre de cette méthode sont validées en utilisant les contraintes heuristiques d'appariement sémantique. La dernière contribution concerne l'amélioration des modèles de rétro-pertinence (Pseudo Relevance Feedback PRF). Étant basée également sur les RDM, notre méthode permet d'intégrer la similarité entre les termes d'expansions et ceux de la requête dans les modèles standards PRF. La validation expérimentale de l'ensemble des contributions apportées dans le cadre de cette thèse est effectuée en utilisant la collection standard TREC 2002/2001 de la langue arabe. / Given the amount of Arabic textual information available on the web, developing effective Information Retrieval Systems (IRS) has become essential to retrieve relevant information. Most of the current Arabic SRIs are based on the bag-of-words representation, where documents are indexed using surface words, roots or stems. Two main drawbacks of the latter representation are the ambiguity of Single Word Terms (SWTs) and term mismatch.The aim of this work is to deal with SWTs ambiguity and term mismatch. Accordingly, we propose four contributions to improve Arabic content representation, indexing, and retrieval. The first contribution consists of representing Arabic documents using Multi-Word Terms (MWTs). The latter is motivated by the fact that MWTs are more precise representational units and less ambiguous than isolated SWTs. Hence, we propose a hybrid method to extract Arabic MWTs, which combines linguistic and statistical filtering of MWT candidates. The linguistic filter uses POS tagging to identify MWTs candidates that fit a set of syntactic patterns and handles the problem of MWTs variation. Then, the statistical filter rank MWT candidate using our proposed association measure that combines contextual information and both termhood and unithood measures. In the second contribution, we explore and evaluate several IR models for ranking documents using both SWTs and MWTs. Additionally, we investigate a wide range of proximity-based IR models for Arabic IR. Then, we introduce a formal condition that IR models should satisfy to deal adequately with term dependencies. The third contribution consists of a method based on Distributed Representation of Word vectors, namely Word Embedding (WE), for Arabic IR. It relies on incorporating WE semantic similarities into existing probabilistic IR models in order to deal with term mismatch. The aim is to allow distinct, but semantically similar terms to contribute to documents scores. The last contribution is a method to incorporate WE similarity into Pseud-Relevance Feedback PRF for Arabic Information Retrieval. The main idea is to select expansion terms using their distribution in the set of top pseudo-relevant documents along with their similarity to the original query terms. The experimental validation of all the proposed contributions is performed using standard Arabic TREC 2002/2001 collection.

Submitted by Vasti Diniz (vastijpa@hotmail.com) on 2017-09-18T11:49:17Z No. of bitstreams: 1 arquivototal.pdf: 2571485 bytes, checksum: 176b4eb5f639864aaef387d41330b286 (MD5) / Made available in DSpace on 2017-09-18T11:49:17Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 2571485 bytes, checksum: 176b4eb5f639864aaef387d41330b286 (MD5) Previous issue date: 2016-03-11 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / After the formulation of general relativity differential geometry has become an increasing important tool in theoretical physics. This is even more clear in the investigation of the so-called embedding space-time theories. In this work we focus our attention in the Cauchy problem. These have played a crucial role in our understanding of the mathematical struc­ture of general relativity and embedding theories. We investigate the similarities and diffe­rences between the two approaches. We also study an extension of the Campbell-Magaard theorem and give two examples of both formalisms. / A geometria diferencial passou a ser uma ferramenta fundamental na fisica com o surgi­mento da relatividade geral. Em particular, destacamos sua importância na investigado das chamadas teorias de imersdo do espaco-tempo. Neste trabalho analisamos dois grandes for­malismos fundamentados de forma direta ou indireta na teoria de imersões: o teorema de Campbell-Magaard e o problema de Cauchy para a relatividade geral. Tendo como princi­pal objetivo tracar um paralelo entre esses dois formalismos, estudamos, nesta dissertacdo, o problema de valor inicial (pvi) para a relatividade geral mostrando que alem de admitir a formulae-do de pvi, a mesma é bem posta. Ademais, aplicamos este formalismo para o caso de uma metrica do tipo Friedmann-Robertson-Walker em (3+1). Estudamos tambem o teorema de Campbell-Magaard e sua extensdo para o espaco-tempo de Einstein e aplicamos este teorema para uma metrica do tipo de Sitter em (2+1).

Resumo: Sistemas dinâmicos híbridos se diferenciam por exibir simultaneamente variados tipos de comportamento dinâmico (contínuo, discreto, eventos discretos) em diferentes partes do sistema. Neste trabalho foram estudados resultados de estabilidade no sentido de Lyapunov para sistemas dinâmicos híbridos gerais, que utilizam uma noção de tempo generalizado, definido em um espaço métrico totalmente ordenado. Mostrou-se que estes sistemas podem ser imersos em sistemas dinâmicos descontínuos definidos em R+, de forma que sejam preservadas suas propriedades qualitativas. Como foco principal, estudou-se resultados de estabilidade para sistemas dinâmicos descontínuos modelados por semigrupos de operadores, em que os estados do sistema pertencem à espaços de Banach. Neste caso, de forma alternativa à teoria clássica de estabilidade, os resultados não utilizam as usuais funções de Lyapunov, sendo portanto mais fáceis de se aplicar, tendo em vista a dificuldade em se encontrar tais funções para muitos sistemas. Além disso, os resultados foram aplicados à uma classe de equações diferenciais com retardo. / Abstract: Hybrid dynamical systems are characterized for showing simultaneously a variety of dynamic behaviors (continuous, discrete, discrete events) in different parts of the System. This work discusses stability results in the Lyapunov sense for general hybrid dynamical systems that use a generalized notion of time, defined in a completely ordered metric space. It has been shown that these systems may be immersed in discontinuous dynamical systems defined in R+, so that their quality properties are preserved. As the main focus, it is studied stability results for discontinuous dynamical systems modeled by semigroup operators, in which the states belong to Banach spaces. In this case, an alternative to the classical theory of stability, the results do not make use of the usual Lyapunov functions, and therefore are easier to apply, in view of the difficulty in finding such functions for many systems. Furthermore, the results were applied to a class of time-delay discontinuous differential equations. / Orientador: Geraldo Nunes Silva / Coorientador: Luís Antônio Fernandes de Oliveira / Banca: Carlos Alberto Raposo da Cunha / Banca: Waldemar Donizete Bastos / Mestre

Made available in DSpace on 2015-05-15T11:46:22Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 1908357 bytes, checksum: 3762912e093a6400855708f530b6cd4d (MD5) Previous issue date: 2013-05-28 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / We study the geometry of surfaces immersed in R4 through the singularities of their families of height functions. Inflection points on the surfaces are shown to be umbilic points from their families of height functions. Furthermore, we see that inflection points of imaginary type are isolated points of the curve --1(0). As a consequence we prove that any dive generic convexly embedded S2 in R4 has inflexion points. / Neste trabalho estudamos a geometria das superfícies em R4 através da variedade canal e das singularidades das famílias de funções altura das superfícies. Provaremos que os pontos de inflexão das superfície são os pontos umbílicos das famílias de funções altura. Além disso, veremos que pontos de inflexão do tipo imaginário serão pontos isolados da curva --1(0). Como uma consequência deste estudo provaremos que qualquer mergulho genérico convexo de S2 em R4 tem pelo menos um ponto de inflexão.

O transplante cardíaco é a última opção terapêutica para pacientes portadores de insuficiência cardíaca grave. Apesar dos avanços na terapia imunossupressora, a rejeição continua sendo o principal obstáculo para o sucesso do transplante. No presente estudo, propõe-se avaliar o perfil de expressão gênica no tecido cardíaco. Com isso, espera-se contribuir para o melhor entendimento do processo de rejeição a nível molecular. Foram analisadas as amostras sequenciais (1, 3 e 6 meses, 1 e 2 anos pós-transplante) de biópsias endomiocárdicas em parafina de 63 indivíduos transplantados cardíacos. O diagnóstico de rejeição humoral foi realizado pela detecção de C3d e C4d do complemento na reação de imuno-histoquímica, e as expressões dos genes protetores (ADIPOR1, ADIPOR2, HMOXO-1, BCL2L1 e VEGF) e genes associados à inflamação (IL-6, TNF?, IFN?, TGF-?, AIF-1, NOS2, ICAM, VCAM e MCP-1); foram avaliadas pela reação em cadeia da polimerase quantitativa em tempo real (qPCR). As frequências de indivíduos positivos para C4d (28,6%) e vasculite (20,0%) foram significantemente maiores em relação ao teste de reatividade de anticorpos realizado nos receptores antes do transplante (6,3%). Houve mudança no perfil de expressão gênica no tecido cardíaco após o transplante, com aumento da expressão dos genes inflamatórios (AIF-1, TNF?, IL-6, NOS2 e VCAM) e diminuição da expressão dos genes protetores (ADIPOR1, ADIPOR2, BCL2L1 e VEGF). Além disso, as expressões dos genes ADIPOR1 e ADIPOR2 foram significantemente menores nos indivíduos positivos para C4d (p<0,001) e gene VEGF (p<0,001) no grupo com vasculite. Houve também uma correlação positiva entre a expressão do gene VEGF e ADIPOR1 (r=0,5688) e ADIPOR2 (r=0,5191). Por outro lado, a expressão aumentada do gene VCAM (p<0,001) foi detectada em todos os tipos de rejeição. Conclui-se que, depois do transplante, o sistema imune do receptor passou a reconhecer os antígenos do órgão transplantado. Com isso, ocorreu uma mudança no perfil de expressão gênica no enxerto cardíaco, caracterizada pela expressão aumentada dos genes inflamatórios e diminuição dos genes protetores. A expressão aumentada do gene VCAM associada à baixa expressão dos genes protetores: ADIPOR1, ADIPOR2, BCL2L1 e VEGF resultaram em maior gravidade da rejeição celular, humoral e vasculopatia do aloenxerto cardíaco. / Heart transplantation is the ultimate treatment for patients with severe heart failure. Despite advances in immunosuppressive therapy, rejection still remains the main obstacle to successful transplant. The purpose of this study is to evaluate the gene expression profile in cardiac tissue. With this we hope to contribute to a better understanding of the rejection process at the molecular level. Sequential samples (1, 3, and 6 months, 1 to 2 years post-transplant) endomyocardial biopsies paraffin of 63 heart transplant subjects were analyzed. The diagnosis of humoral rejection was performed by detection of C3d and C4d complement in immunohistochemistry and the expression of protective genes (ADIPOR1, ADIPOR2, HMOXO-1, and VEGF BCL2L1) and genes associated with inflammation (IL-6,TNF?, IFN?, TGF-?, AIF-1, NOS2, ICAM, VCAM and MCP-1); were evaluated by quantitative real-time polymerase chain reaction (qPCR). The frequency of individuals positive for C4d (28.6%) and vasculitis (20.0%) were significantly higher compared to antibodies reactivity test conducted in the recipients before transplantation (6.3%). There was a change in the gene expression profile in cardiac tissue after transplantation, with increased expression of inflammatory genes (AIF-1,TNF?, IL-6, NOS2, and VCAM) and a decreased expression of protective genes (ADIPOR1, ADIPOR2, BCL2L1, and VEGF). Furthermore, the expressions of ADIPOR1 and ADIPOR2 genes were significantly lower in C4d positive individuals (p<0,001), and VEGF (p<0,001) in the group with vasculitis. There was also a positive correlation between VEGF expression and ADIPOR1 (r=0.5688) and DIPOR2 (r=0.5191). Moreover, increased expression of VCAM (p<0,001) was detected in all types of rejection. We conclude that, after transplantation, the recipient\'s immune system began to recognize the antigens of the transplanted organ. Thus, a change in gene expression profile in cardiac graft is characterized by increased inflammatory genes and decreased expression of protective genes. Increased VCAM gene associated with lower expression of protective genes expression: ADIPOR1, ADIPOR2, BCL2L1 and VEGF resulted in increased severity of cellular and humoral rejection and cardiac allograft vasculopathy.

INTRODUÇÃO: A punção aspirativa por agulha fina guiada por ecoendoscopia (PAAF-EE) é método consagrado para o diagnóstico de lesões sólidas de pâncreas. A aplicação de pressão negativa (PN) durante a PAAF-EE tem objetivo de manter o tecido junto à ponta cortante da agulha durante sua passagem através da lesão, resultando no desprendimento e retenção de células. A capilaridade (CP) é outra técnica que utiliza a retirada gradual do estilete durante a PAAF-EE e parece ser igualmente eficaz, além de evitar aspiração indesejada de hemácias e coágulos. OBJETIVO: Analisar o rendimento da ecopunção de lesões sólidas pancreáticas utilizando PN versus (vs) CP e os aspectos citológicos qualitativos e semiquantitativos da celularidade e contaminação por sangue dos grupos estudados. MÉTODO: No estudo, foram randomizados 50 pacientes com lesões sólidas de pâncreas para realização de PAAF-EE. Cada paciente foi submetido a quatro punções, intercalando as duas técnicas. A ordem das punções foi randomizada por computador. O grupo A (PN, CP, PN, CP) e o grupo B (CP, PN, CP, PN). RESULTADOS: As sensibilidade, especificidade e acurácia da PN vs CP foram 95,2% vs 92,3%; 100% vs 100%; 95,7% vs 93%, respectivamente. Quando foi considerada a associação das duas técnicas (PN+CP), os resultados da sensibilidade, especificidade e acurácia foram 95,6%, 100% e 96%. O valor preditivo positivo (VPP) foi alto para todas técnicas (100%) e os valores preditivos negativos (VPN) para PN, CP e associação dos métodos foram de 66,7% vs 57,1% vs 66,7%, respectivamente. Não houve diferença estatística no rendimento diagnóstico entre a PN e CP: 88% vs 80% (p=0,344), entretanto as técnicas somadas (PN+CP) foi superior à CP isolada: 94% vs 80% (p=0,016). A avaliação citológica qualitativa e semi-quantitativa de celularidade e contaminação por sangue no esfregaço e emblocado pelas técnicas de PN e CP não apresentou diferença estatística (p>0,05). CONCLUSÃO: O rendimento diagnóstico da PAAF-EE pela técnica de PN versus CP de massas sólidas de pâncreas não apresenta diferença estatística, entretanto a utilização das técnicas combinadas é superior, particularmente em relação à CP isolada. A celularidade e contaminação por sangue é semelhante independente da técnica utilizada (PN ou CP) ou do tipo de preparação citológica (esfregaço ou emblocado) / BACKGROUND: Endoscopic ultrasound fine needle aspiration (EUS-FNA) is an established diagnostic method for solid pancreatic masses. Application of negative pressure (NP) during EUS-FNA has the purpose of holding tissue against the cutting edge of the needle as it is moved through the lesion leading to cell detachment and drawing them up. Capillary (CP) sampling is another technique that removes the needle stylet gradually and continuously during the EUS-FNA and seems to have comparable results with less blood and clot aspiration. AIM: Analyze diagnostic yield of EUS-FNA of pancreatic solid mass using NP versus CP and evaluate cytological qualitative and semi-quantitative features of cellularity and blood contamination of each group. METHOD: A total of 50 patients were randomized for NP or CP EUSFNA with a standard 22G needle. Each patient had four passes using NP and CP in an alternate fashion. Computer randomization generated two groups. Group A (NP, CP, NP, CP) and group B (CP, NP, CP, NP). RESULTS: Diagnostic yield of NP versus CP had no statistical difference (p=0.344). Both techniques (NP+CP) was superior to CP alone (p=0.016), but not to NP alone (p=0.25). Sensitivity, specificity and accuracy of NP vs CP were 95.2% vs 92.3%; 100% vs 100%; 95.7% vs 93%, respectively. Results for sensitivity, specificity and accuracy of both NP+CP were 95.6%, 100% and 96%. Positive predictive value was high for all techniques (100%) and Negative predictive value for NP, CP and both techniques were 66.7%, 57.1% and 66.7%, respectively. NP and CP smears and cell blocks had no statistical differences in qualitative and semi-quantitative citological evaluation for cellularity and bloodiness (p > 0.05). CONCLUSIONS: NP versus CP in EUS-FNA of solid pancreatic mass yielded similar overall outcomes in terms of diagnostic performance, however the combination of both techniques is superior, particularly compared to CP alone. Cellularity and bloodiness are equivalent no matter which technique (NP or CP) or type of cytological preparation (smear or cell block) is used

La constante amélioration des technologies silicium permet aux transistors bipolaires à hétérojonction (HBT) SiGeC (Silicium-Germanium : Carbone) de concurrencer les composants III-V pour les applications millimétriques et sous-THz (jusqu’à 300GHz). Le cycle de développement de la technologie (caractérisation-modélisation-conception-fabrication) nécessite plusieurs itérations, entraînant des coûts élevés. De plus, les méthodologies de mesure doivent être réévaluées et ajustées pour adresser des fréquences plus élevées. Afin de réduire le nombre d’itérations et de permettre la montée en fréquence de la mesure, un travail de fond sur la première étape, la caractérisation, s’avère indispensable.Pour répondre à cette exigence, une description et une étude des instruments de mesure (VNA) est réalisée dans un premier temps. Un état de l’art des méthodes de calibrage permet de choisir la solution la plus pertinente pour la calibration sur puce valable dans la gamme de fréquences sous-THz. Ensuite, après avoir relevé plusieurs défauts dans la méthode choisie (à savoir la méthode Thru-Reflect-Line : TRL), des solutions sont proposées concernant la modification des calculs des coefficients d’erreur et également en modifiant les standards utilisés durant le calibrage. Finalement, une étude sur les méthodes d’épluchage est réalisée. Une amélioration est proposée par la modification de deux standards évitant le principal problème de l’état de l’art, la surcompensation des composants parasites. / The continuous improvement in Silicon technologies allows SiGeC (Silicon-Germanium-Carbon) heterojunction bipolar transistors (HBT) to compete with III-V components for millimeter wave and sub-THz (below 300GHz) applications. The technology development cycle (characterization, modeling, design and fabrication) needs several iterations resulting in high costs. Furthermore, the measurement methodologies need to be re-assessed and modified to address higher measurement frequencies. In order to reduce the number of iterations and to allow reliable measurement in the sub-THz band, the characterization procedure has been revisited.First, a description and investigation of the measurement instrument (VNA) has been made. After exploring all possible calibration methods, the best candidate for an “on-wafer” calibration for the sub-THz frequency range has been selected. Then, after analyzing the limits of the chosen calibration method (Thru-Reflect-Line: TRL), workarounds are proposed, by modification of the errors coefficients calculation and by changing the standards used during the calibration process. At last, a study concerning the de-embedding methods is carried out. It is shown, that using two new standards helps to reduce the over-compensation of parasitic components.

以往的音樂資訊檢索方法多使用歌詞、曲風、演奏的樂器或一段音頻訊號來當作查詢的媒介，然而，在某些情況下，使用者沒有辦法清楚描述他們想要尋找的歌曲，如：情境式的音樂檢索。本論文提出了一種基於圖像的情境式音樂資訊檢索方法，可以透過輸入圖片來找尋相應的音樂。此方法中我們使用了卷積神經網絡（Convolutional Neural Network）技術來處理圖片，將其轉為低維度的表示法。為了將異質性的多媒體訊息映射到同一個向量空間，資訊網路表示法學習（Network Embedding）技術也被使用，如此一來，可以使用距離計算找回和輸入圖片有關的多媒體訊息。我們相信這樣的方法可以改善異質性資訊間的隔閡（Heterogeneous Gap），也就是指不同種類的多媒體檔案之間無法互相轉換或詮釋。在實驗與評估方面，首先利用從歌詞與歌名得到的關鍵字來搜尋大量圖片當作訓練資料集，接著實作提出的檢索方法，並針對實驗結果做評估。除了對此方法的有效性做測試外，使用者的回饋也顯示此檢索方法和其他方法相比是有效的。同時我們也實作了一個網路原型，使用者可以上傳圖片並得到檢索後的歌曲，實際的使用案例也將在本論文中被展示與介紹。 / Listening to music is indispensable to everyone. Music information retrieval systems help users find their favorite music. A common scenario of music information retrieval systems is to search songs based on user's query. Most existing methods use descriptions (e.g., genre, instrument and lyric) or audio signal of music as the query; then the songs related to the query will be retrieved. The limitation of this scenario is that users might be difficult to describe what they really want to search for. In this paper, we propose a novel method, called ``image2song,'' which allows users to input an image to retrieve the related songs. The proposed method consists of three modules: convolutional neural network (CNN) module, network embedding module, and similarity calculation module. For the processing of the images, in our work the CNN is adopted to learn the representations for images. To map each entity (e.g., image, song, and keyword) into a same embedding space, the heterogeneous representation is learned by network embedding algorithm from the information graph. This method is flexible because it is easy to join other types of multimedia data into the information graph. In similarity calculation module, the Euclidean distance and cosine distance is used as our criterion to compare the similarity. Then we can retrieve the most relevant songs according to the similarity calculation. The experimental results show that the proposed method has a good performance. Furthermore, we also build an online image-based music information retrieval prototype system, which can showcase some examples of our experiments.

La segmentation des données par la langue et le thème des tests psycholinguistiques devient de plus en plus un obstacle important à la généralisation des modèles de prédiction. Cela limite notre capacité à comprendre le cœur du dysfonctionnement linguistique et cognitif, car les modèles sont surajustés pour les détails d'une langue ou d'un sujet particulier. Dans ce travail, nous étudions les approches potentielles pour surmonter ces limitations. Nous discutons des propriétés de divers modèles de plonjement de mots FastText pour l'anglais et le français et proposons un ensemble des caractéristiques, dérivées de ces propriétés. Nous montrons que malgré les différences dans les langues et les algorithmes de plonjement, un ensemble universel de caractéristiques de vecteurs de mots indépendantes de la langage est capable de capturer le dysfonctionnement cognitif. Nous soutenons que dans le contexte de données rares, les caractéristiques de vecteur de mots fabriquées à la main sont une alternative raisonnable pour l'apprentissage des caractéristiques, ce qui nous permet de généraliser sur les limites de la langue et du sujet. / Data segmentation by the language and the topic of psycholinguistic tests increasingly becomes a significant obstacle for generalization of predicting models. It limits our ability to understand the core of linguistic and cognitive dysfunction because the models overfit the details of a particular language or topic. In this work, we study potential approaches to overcome such limitations. We discuss the properties of various FastText word embedding models for English and French and propose a set of features derived from these properties. We show that despite the differences in the languages and the embedding algorithms, a universal language-agnostic set of word-vector features can capture cognitive dysfunction. We argue that in the context of scarce data, the hand-crafted word-vector features is a reasonable alternative for feature learning, which allows us to generalize over the language and topic boundaries.