A multimodal network (MMN) is a novel mathematical construct that captures the structure of biological networks, computational network models, and relationships from biological databases. An MMN subsumes the structure of graphs and hypergraphs, either undirected or directed. Formally, an MMN is a triple (V,E,M) where V is a set of vertices, E is a set of modal hyperedges, and M is a set of modes. A modal hyperedge e=(T,H,A,m) in E is an ordered 4-tuple, in which T,H,A are subsets of V and m is an element of M. The sets T, H, and A are the tail, head, and associate of e, while m is its mode. In the context of biology, each vertex is a biological entity, each hyperedge is a relationship, and each mode is a type of relationship (e.g., 'forms complex' and 'is a'). Within the space of multimodal networks, structural operations such as union, intersection, hyperedge contraction, subnetwork selection, and graph or hypergraph projections can be performed. A denotational semantics approach is used to specify the semantics of each hyperedge in MMN in terms of interaction among its vertices. This is done by mapping each hyperedge e to a hyperedge code algo:V(e), an algorithm that details how the vertices in V(e) get used and updated. A semantic MMN-based model is a function of a given schedule of evaluation of hyperedge codes and the current state of the model, a set of vertex-value pairs. An MMN-based computational system is implemented as a proof of concept to determine empirically the benefits of having it. This system consists of an MMN database populated by data from various biological databases, MMN operators implemented as database functions, graph operations implemented in C++ using LEDA, and mmnsh, a shell scripting language that provides a consistent interface to both data and operators. It is demonstrated that computational network models may enrich the MMN database and MMN data may be used as input to other computational tools and environments. A simulator is developed to compute from an initial state and a schedule of hyperedge codes the resulting state of a semantic MMN model. / Ph. D.

CoordenaÃÃo de AperfeiÃoamento de Pessoal de NÃvel Superior / O transporte pÃblico urbano à uma alternativa de soluÃÃo para os impactos negativos do modelo atual de transporte, e para viabilizar uma oferta de qualidade, sÃo necessÃrias ferramentas que possibilitem um planejamento eficiente dos sistemas de maneira integrada. Devido a suas caracterÃsticas especÃficas, redes multimodais integradas de transporte pÃblico apresentam problemas associados à modelagem dos tipos de integraÃÃo e da escolha de rota, os quais, em vÃrias perspectivas, sÃo mais complexos do que os encontrados em redes viÃrias. Desta forma, esta dissertaÃÃo realiza uma anÃlise conceitual e aplicada da modelagem de redes multimodais integradas de transporte pÃblico urbano, utilizando a Zona Oeste da RegiÃo Metropolitana de Fortaleza por meio do programa TransCAD, e verificando-se a eficiÃncia dos modelos combinados de escolha de modos e rotas quanto à inserÃÃo do carÃter aleatÃrio e à representaÃÃo dos efeitos do congestionamento. Ao considerar a alocaÃÃo da demanda por meio de um mÃtodo de estratÃgias, observou-se que houve reduÃÃo das parcelas de tempo relacionadas aos transbordos, confirmando sua caracterÃstica de minimizar o tempo gasto nas paradas pelos usuÃrios. A adoÃÃo da restriÃÃo da capacidade provocou aumento significativo no custo generalizado, alÃm de apresentar um maior tempo de caminhada. TambÃm houve impacto sobre a dispersÃo da demanda na rede, apresentando maior variaÃÃo de rotas carregadas. Ao considerar efeitos estocÃsticos, observou-se aumento em parcelas do tempo de viagem, alÃm da distÃncia percorrida quando embarcados. Por fim, conclui-se que os mÃtodos de estratÃgia e os mÃtodos de equilÃbrio sÃo bastante distintos quanto aos resultados da alocaÃÃo, com diferenÃas significativas nas parcelas de tempo relacionadas ao transbordo. Percebe-se que, apesar de o EquilÃbrio EstocÃstico do UsuÃrio ser considerado pela comunidade cientÃfica como o mÃtodo do estado da arte, ao ser submetido a certas condiÃÃes, pode apresentar resultados semelhantes a um mÃtodo bastante simplista como o Tudo ou Nada. Assim, nota-se que quando nÃo hà domÃnio de cada etapa que està sendo executada, um mÃtodo pode se comportar de maneira inesperada, comprometendo os resultados de todo o processo, e consequentemente, nÃo serà uma ferramenta eficaz para apoiar projetos de intervenÃÃes nos sistemas de transportes. / The urban public transportation is an alternative solution to the negative impacts of the current transportation model, and, to enable a quality supply of transportation services, tools are required to allow efficient systems planning in an integrated way. Due to their specific characteristics, integrated multimodal transit networks have problems related to the modeling of the types of integration and route choice, which, from many perspectives, are more complex than those found in road networks. Thus, in this paper there is a conceptual and applied analysis about integrated multimodal transit networks modeling, using the Western Area of the Metropolitan Region of Fortaleza with software TransCAD, and checking the efficiency of the combined models of modal and route choice and the insertion of random character and representation of the effects of congestion. In the assignment by a strategy method, it is possible to observe that there was a reduction of time related to transfers, confirming its characteristic to minimize the time spent by users on the stops. The adoption of the capacity constraint caused a significant increase in generalized cost and in walk time. The adoption of the capacity constraint also shown an impact on the dispersion of demand on the network, presenting great variation of loaded routes. Adopting stochastic effects, there was an increase in parcels of travel time and in in-vehicle distance. Finally, we can conclude that the strategy methods and the equilibrium methods are quite different about the assignment results, with significant differences in the time parcels related to transfers. It is noticed that, although the scientific community consider the Stochastic User Equilibrium as the state of the art method, when it is subjected to certain situations, it can present the same results as a rather simplistic method like All or Nothing. Therefore, when there is no command of each executed step, an assignment method may behave unexpectedly, compromising the results of the whole process, and consequently, it will not be an effective tool to support intervention projects in transportation systems.

Neural Networks are at the core of computer vision solutions for various applications. With the advent of deep neural networks Facial Expression Recognition (FER) has been a very ineluctable and challenging task in the field of computer vision. Micro-expressions (ME) have been quite prominently used in security, psychotherapy, neuroscience and have a wide role in several related disciplines. However, due to the subtle movements of facial muscles, the micro-expressions are difficult to detect and identify. Due to the above, emotion detection and classification have always been hot research topics. The recently adopted networks to train FERs are yet to focus on issues caused due to overfitting, effectuated by insufficient data for training and expression unrelated variations like gender bias, face occlusions and others. Association of FER with the Speech Emotion Recognition (SER) triggered the development of multimodal neural networks for emotion classification in which the application of sensors played a significant role as they substantially increased the accuracy by providing high quality inputs, further elevating the efficiency of the system. This thesis relates to the exploration of different principles behind application of deep neural networks with a strong focus towards Convolutional Neural Networks (CNN) and Generative Adversarial Networks (GAN) in regards to their applications to emotion recognition. A Motion Magnification algorithm for ME's detection and classification was implemented for applications requiring near real-time computations. A new and improved architecture using a Multimodal Network was implemented. In addition to the motion magnification technique for emotion classification and extraction, the Multimodal algorithm takes the audio-visual cues as inputs and reads the MEs on the real face of the participant. This feature of the above architecture can be deployed while administering interviews, or supervising ICU patients in hospitals, in the auto industry, and many others. The real-time emotion classifier based on state-of-the-art Image-Avatar Animation model was tested on simulated subjects. The salient features of the real-face are mapped on avatars that are build with a 3D scene generation platform. In pursuit of the goal of emotion classification, the Image Animation model outperforms all baselines and prior works. Extensive tests and results obtained demonstrate the validity of the approach.

There exist a large number of rare genetic diseases in humans. Our knowledge of the specific gene variants whose presence in the genome of a person predisposes them towards developing a disease, called gene associations, is incomplete. Computational tools which can predict genes which may be associated with a rare disease have great utility in healthcare. However, a majority of existing prediction algorithms require a set of already known "seed genes'' to further discover novel associations for a disease. This drawback becomes more serious for rare genetic diseases, since a large proportion do not have any known gene associations. In this work, we develop an approach for disease-gene association prediction that overcomes the reliance on seed genes. Our approach uses the similarity of the observable biological characteristics of diseases (i.e., phenotypes) along with a global map of direct and indirect human protein interactions, to transfer associations from diseases whose gene associations have been discovered to diseases with no known gene associations. We formulate disease-gene association prediction over a multimodal network of diseases and genes, and develop an approach based on graph convolutional networks. We show how our model design considerations impact prediction performance. We demonstrate that our approach outperforms simpler graph machine learning and traditional machine learning approaches, as well as a competitive network propagation based approach for the task of predicting disease-gene associations. / Master of Science / There exist a large number of rare genetic diseases in humans. Our knowledge of the specific gene variants whose presence in the genome of a person predisposes them towards developing a disease, called gene associations, is incomplete. Computational tools which can predict genes which may be associated with a rare disease have great utility in healthcare. However, a majority of existing prediction algorithms require a set of already known "seed genes'' to further discover novel associations for a disease. This drawback becomes more serious for rare genetic diseases, since a large proportion do not have any known gene associations. In this work, we develop an approach for disease-gene association prediction that overcomes the reliance on seed genes. Our approach uses the similarity of the observable biological characteristics of diseases (i.e. disease phenotypes) along with a global map of direct and indirect human protein interactions, to transfer gene associations from diseases whose gene associations have been discovered, to diseases with no known associations. We implement an approach based on the field of graph machine learning, namely graph convolutional networks, to predict the genes associated with rare genetic diseases. We show how our predictor performs, compared to other approaches, and analyze some of the choices made in the design of the predictor, along with some properties of the outputs of our predictor.