Abstract
This paper proposes and reports the development of an open source solution for the integrated management of Infrastructure as a Service (IaaS) cloud computing resources, through the use of a common API taxonomy, to incorporate open source and proprietary platforms. This research included two surveys on open source IaaS platforms (OpenNebula, OpenStack and CloudStack) and a proprietary platform (Parallels Automation for Cloud Infrastructure - PACI) as well as on IaaS abstraction solutions (jClouds, Libcloud and Deltacloud), followed by a thorough comparison to determine the best approach. The adopted implementation reuses the Apache Deltacloud open source abstraction framework, which relies on the development of software driver modules to interface with different IaaS platforms, and involved the development of a new Deltacloud driver for PACI. The resulting interoperable solution successfully incorporates OpenNebula, OpenStack (reuses pre-existing drivers) and PACI (includes the developed Deltacloud PACI driver) nodes and provides a Web dashboard and a Representational State Transfer (REST) interface library. The results of the exchanged data payload and time response tests performed are presented and discussed. The conclusions show that open source abstraction tools like Deltacloud allow the modular and integrated management of IaaS platforms (open source and proprietary), introduce relevant time and negligible data overheads and, as a result, can be adopted by Small and Medium-sized Enterprise (SME) cloud providers to circumvent the vendor lock-in problem whenever service response time is not critical.

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
This paper presents a set of methods and techniques to remove inefficiencies in a data analysis application used in searches by the ATLAS Experiment at the Large Hadron Collider. Profiling scientific code helped to pinpoint design and runtime inefficiencies, the former due to coding and data structure design. The data analysis code used by groups doing searches in the ATLAS Experiment contributed to clearly identify some of these inefficiencies and to give suggestions on how to prevent and overcome those common situations in scientific code to improve the efficient use of available computational resources in a parallel homogeneous platform.