Abstract
Transfer Learning is a paradigm in machine learning to solve a target problem by reusing the learning with minor modifications from a different but related source problem. In this paper we propose a novel feature transference approach, especially when the source and the target problems are drawn from different distributions. We use deep neural networks to transfer either low or middle or higher-layer features for a machine trained in either unsupervised or supervised way. Applying this feature transference approach on Convolutional Neural Network and Stacked Denoising Autoencoder on four different datasets, we achieve lower classification error rate with significant reduction in computation time with lower-layer features trained in supervised way and higher-layer features trained in unsupervised way for classifying images of uppercase and lowercase letters dataset. © 2014 Springer International Publishing Switzerland.

Abstract
In predictive modeling tasks, knowledge about the training examples is neither fully complete nor totally incomplete. Unlike semisupervised learning, where one either has perfect knowledge about the label of the point or is completely ignorant about it, here we address a setting where, for each example, we only possess partial information about the label. Each example is described using two (or more) different feature sets or views, where neither are necessarily observed for a given example. If a single view is observed, then the class is only due to that feature set; if more views are present, the observed class label is the maximum of the values corresponding to the individual views. After formalizing this new learning concept, we propose two new learning methodologies that are adapted to this learning paradigm. We also compare their instantiation in experiments with different base models and with conventional methods. The experimental results made both on real and synthetic data sets verify the usefulness of the proposed approaches.

Abstract
Wireless visual sensor networks provide valuable informa-tion for many monitoring and control applications. Some-times, a set of targets need to be monitored by deployed visual sensors. For those networks, however, some active vi-sual sources may fail, potentially degrading the application monitoring quality when targets become uncovered. More-over, some applications may need different perspectives of the same target. As visual sensors will be used to moni-tor a set of targets, a high level of monitoring redundancy may be required and an effective way to achieve it is as-suring that targets are being concurrently viewed by more than one visual sensor. We propose a centralized greedy algorithm to enhance redundancy in wireless visual sensor networks when visual sensors with adjustable orientations are deployed. Additionally, as some targets may be more critical for the application, we propose a priority-based con-figuration of the sensors' poses in order to find an optimized configuration for the visual sensors.

Abstract
Crete Island has significant natural resources when it comes to wind and solar energy. Likewise other European territories, renewable sources already are being explored for power production. Currently, a large amount of wind energy on Crete is curtailed during certain daily periods as a result of reduced demand and minimum operating levels of thermal generators. Reducing curtailment losses requires additional sources of flexibility in the grid, and electric energy storage is one of them. This paper address wind generation losses minimization through the storage of wind energy surplus. Sodium Sulfur (NaS) battery modeling is used in this study and an energy time-shift storage scheme is implemented to assess the overall storage system performance. The obtained results are supported on real data of renewable resources (wind and solar), conventional power production and demand of Crete Island in 2011. Conclusions are duly drawn.

Abstract
The analysis of the functional properties (ferroelectric, dielectric, and piezoelectric) of chemical solution deposited thin films of the lead-free (Bi0.5Na0.5)(1-x)BaxTiO3 (BNBT) solid solution prepared from solution precursors with and without Na+ and Bi3+ excesses has been performed in this work. At room temperature a nonergodic relaxor ferroelectric state has been found. The switched polarization of the films is not stable at room temperature, poor remnant polarization, associated with an enhancement of the induced domains randomization produced by the films constraints. The depolarization temperature for the switched polarization allowed us to build up a tentative phase diagram for these BNBT films. Both the better functional properties and the agreement of the depolarization temperature with the freezing temperature of the relaxor Volger-Fulcher behavior permit to locate the center of the morphotropic phase boundary region close to x=0.055 in the stoichiometric films and x=0.10 for the films with Na+ and Bi3+ excesses. Based on these results, the possible applications of these films are discussed.

Abstract
Web search engines are powerful tools used to satisfy specific information needs on the web. Their purpose is to maximize user satisfaction when performing this task. Although there are other sources of evidence, besides text, to characterize document relevance for a specific need, current search engines do not allow users to explore these features when posing a query. Search engine queries are based almost exclusively on keywords. It is possible to improve user satisfaction if HTML tags and document metadata are available to users at query time. In this paper we present Xearch, a meta-search system that wraps public search engines in a framework that improves both the expressiveness of the language available for the user to specify information needs and the control over the answer format. Xearch converts HTML pages to a specific XML Schema, covering text and metadata derived from HTML. User queries can then be specified through a set of keywords but also explore documents' HTML tags and metadata. Results from our experimental evaluation confirm that it is possible to improve the answer quality with this framework.