Search

In this paper, a convolutional neural network (CNN)-based deep learning (DL) architecture for the solution of an electromagnetic inverse problem related to imaging of the shape of the perfectly electric conducting (PEC) rough surfaces is addressed. The rough surface is illuminated by a plane wave and scattered field data is obtained synthetically through the numerical solution of surface integral equations. An effective CNN-DL architecture is implemented through the modelling of the rough surface variation in terms of convenient spline type base functions. The algorithm is numerically tested with various scenarios including amplitude only data and shown that it is very effective and useful. Publisher's Version Q2 WOS:000836577000001

In the food and agricultural industries, it is crucial to identify and to choose correct sunflower seeds that meet specific requirements. Deep learning and computer vision methods can help identify sunflower seeds. In this study, a computer vision system was proposed, trained, and tested to identify four varieties of sunflower seeds using deep learning methodology and a regular color camera. Image acquisition was carried out under controlled illumination conditions. An image segmentation procedure was employed to reduce the workload in obtaining training images required for training deep convolutional neural network models. Three deep learning architectures, namely AlexNet, GoogleNet, and ResNet, were investigated for identifying sunflower seeds in this study. Different solver types were also evaluated to determine the best deep learning model in terms of both accuracy and training time. About 4800 sunflower seeds were inspected individually for training and testing. The highest classification accuracy (95%) was succeeded with the GoogleNet algorithm.

International audience; The huge amount of data currently produced by modern Earth Observation (EO) missions has allowed for the design of advanced machine learning techniques able to support complex Land Use/Land Cover (LULC) mapping tasks. The Copernicus programme developed by the European Space Agency provides, with missions such as Sentinel-1 (S1) and Sentinel-2 (S2), radar and optical (multi-spectral) imagery, respectively, at 10 m spatial resolution with revisit time around 5 days. Such high temporal resolution allows to collect Satellite Image Time Series (SITS) that support a plethora of Earth surface monitoring tasks. How to effectively combine the complementary information provided by such sensors remains an open problem in the remote sensing field. In this work, we propose a deep learning architecture to combine information coming from S1 and S2 time series, namely TWINNS (TWIn Neural Networks for Sentinel data), able to discover spatial and temporal dependencies in both types of SITS. The proposed architecture is devised to boost the land cover classification task by leveraging two levels of complementarity, i.e., the interplay between radar and optical SITS as well as the synergy between spatial and temporal dependencies. Experiments carried out on two study sites characterized by different land cover characteristics (i.e., the Koumbia site in Burkina Faso and Reunion Island, a overseas department of France in the Indian Ocean), demonstrate the significance of our proposal.

Wind Energy generation depends on the existence of wind, a meteorological phenomena intermittent by nature, with the consequence of generating uncertainty on the availability of wind energy in the future. The grid stability processes require continuous forecasting of wind energy generated. Forecasting wind energy can be performed either by using weather forecast data or by projecting (or regressing) the past time-series data observations into the future. This last method is the statistical or time series approach. Wind Time Series show non-linearity and non-stationarity properties, and these two properties increase the complexity of the forecasting task using statistical methodologies. In this paper we explore the use of deep learning techniques, which can represent non-linearity, to the wind speed prediction using the largest public wind dataset available, the Wind Toolkit from the National Renewable Laboratory of the US. Several deep network architectures like Multi Layer Perceptrons, Convolutional Networks or Recurrent Networks have been tested on the 126,692 wind-sites and with the results obtained valuable comparisons and conclusions have been obtained. The distribution of the wind sites across the North American Geography has allowed to include in the analysis relationships between terrain, wind forecast complexity and deep methods. With the developed testing workbench and with the availability of the Barcelona Supercomputing Center new architectures are being developed. This work concludes with the feasibility of deep learning architectures for the wind and energy forecasting. We thank the Barcelona Supercomputing Center for the extensive use of their infrastructure in this project, and the United States National Renewable Energy Laboratory (NREL) for the use of their wind energy datasets. This work is partially supported by the Joint Study Agreement no. W156463 under the IBM/BSC Deep Learning Center agreement, by the Spanish Government through Programa Severo Ochoa (SEV-2015-0493), by the Spanish Ministry of Science and Technology through TIN2015-65316-P project, and by the Generalitat de Catalunya (contracts 2014-SGR-1051)

In response to the new challenges in the design and operation of communication networks, and taking inspiration from how living beings deal with complexity and scalability, in this paper we introduce an innovative system concept called COgnition-BAsed NETworkS (COBANETS). The proposed approach develops around the systematic application of advanced machine learning techniques and, in particular, unsupervised deep learning and probabilistic generative models for system-wide learning, modeling, optimization, and data representation. Moreover, in COBANETS, we propose to combine this learning architecture with the emerging network virtualization paradigms, which make it possible to actuate automatic optimization and reconfiguration strategies at the system level, thus fully unleashing the potential of the learning approach. Compared with the past and current research efforts in this area, the technical approach outlined in this paper is deeply interdisciplinary and more comprehensive, calling for the synergic combination of expertise of computer scientists, communications and networking engineers, and cognitive scientists, with the ultimate aim of breaking new ground through a profound rethinking of how the modern understanding of cognition can be used in the management and optimization of telecommunication networks.

Date of Conference: 28 Nov.-1 Dec. 2016 Conference name: 12th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS), 2016 An existing deep learning architecture has been adapted to solve the detection problem in camera-based tracking for augmented reality (AR). A known target, in this case a planar object, is rendered under various viewing conditions including varying orientation, scale, illumination and sensor noise. The resulting corpus is used to train a convolutional neural network to match given patches in an incoming image. The results show comparable or better performance compared to state of art methods. Timing performance of the detector needs improvement but when considered in conjunction with the robust pose estimation process promising results are shown. © 2016 IEEE.

Doutoramento em Engenharia Informática Submitted by Daisy Tavares (daisytavares@ua.pt) on 2014-11-18T11:33:56Z No. of bitstreams: 1 Tese.pdf: 3756875 bytes, checksum: cc9cf426a235f1132f15c6e9159871bb (MD5) Made available in DSpace on 2014-11-18T11:33:56Z (GMT). No. of bitstreams: 1 Tese.pdf: 3756875 bytes, checksum: cc9cf426a235f1132f15c6e9159871bb (MD5) Previous issue date: 2014

Doutoramento em Engenharia Informática This thesis addresses the problem of word learning in computational agents. The motivation behind this work lies in the need to support language-based communication between service robots and their human users, as well as grounded reasoning using symbols relevant for the assigned tasks. The research focuses on the problem of grounding human vocabulary in robotic agent’s sensori-motor perception. Words have to be grounded in bodily experiences, which emphasizes the role of appropriate embodiments. On the other hand, language is a cultural product created and acquired through social interactions. This emphasizes the role of society as a source of linguistic input. Taking these aspects into account, an experimental scenario is set up where a human instructor teaches a robotic agent the names of the objects present in a visually shared environment. The agent grounds the names of these objects in visual perception. Word learning is an open-ended problem. Therefore, the learning architecture of the agent will have to be able to acquire words and categories in an openended manner. In this work, four learning architectures were designed that can be used by robotic agents for long-term and open-ended word and category acquisition. The learning methods used in these architectures are designed for incrementally scaling-up to larger sets of words and categories. A novel experimental evaluation methodology, that takes into account the openended nature of word learning, is proposed and applied. This methodology is based on the realization that a robot’s vocabulary will be limited by its discriminatory capacity which, in turn, depends on its sensors and perceptual capabilities. An extensive set of systematic experiments, in multiple experimental settings, was carried out to thoroughly evaluate the described learning approaches. The results indicate that all approaches were able to incrementally acquire new words and categories. Although some of the approaches could not scale-up to larger vocabularies, one approach was shown to learn up to 293 categories, with potential for learning many more. Esta tese aborda o problema da aprendizagem de palavras em agentes computacionais. A motivação por trás deste trabalho reside na necessidade de suportar a comunicação baseada em linguagem entre os robôs de serviço e os seus utilizadores humanos, bem como suportar o raciocínio baseado em símbolos que sejam relevantes no contexto das tarefas atribuídas e cujo significado seja definido com base na experiência perceptiva. Mais especificamente, o foco da investigação é o problema de estabelecer o significado das palavras na percepção do robô através da interacção homemrobô. A definição do significado das palavras com base em experiências perceptuais e perceptuo-motoras enfatiza o papel da configuração física e perceptuomotora do robô. Entretanto, a língua é um produto cultural criado e adquirido através de interacções sociais. Isso destaca o papel da sociedade como fonte linguística. Tendo em conta estes aspectos, um cenário experimental foi definido no qual um instrutor humano ensina a um agente robótico os nomes dos objectos presentes num ambiente visualmente partilhado. O agente associa os nomes desses objectos à sua percepção visual desses objectos. A aprendizagem de palavras é um problema sem objectivo pré-estabelecido. Nós adquirimos novas palavras ao longo das nossas vidas. Assim, a arquitectura de aprendizagem do agente deve poder adquirir palavras e categorias de uma forma semelhante. Neste trabalho foram concebidas quatro arquitecturas de aprendizagem que podem ser usadas por agentes robóticos para aprendizagem e aquisição de novas palavras e categorias, incrementalmente. Os métodos de aprendizagem utilizados nestas arquitecturas foram projectados para funcionar de forma incremental, acumulando um conjunto cada vez maior de palavras e categorias. É proposta e aplicada uma nova metodologia da avaliação experimental que leva em conta a natureza aberta e incremental da aprendizagem de palavras. Esta metodologia leva em consideração a constatação de que o vocabulário de um robô será limitado pela sua capacidade de discriminação, a qual, por sua vez, depende dos seus sensores e capacidades perceptuais. Foi realizado um extenso conjunto de experiências sistemáticas em múltiplas situações experimentais, para avaliar cuidadosamente estas abordagens de aprendizagem. Os resultados indicam que todas as abordagens foram capazes de adquirir novas palavras e categorias incrementalmente. Embora em algumas das abordagens não tenha sido possível atingir vocabulários maiores, verificou-se que uma das abordagens conseguiu aprender até 293 categorias, com potencial para aprender muitas mais.