Automation of vineyards cultivation necessitates for mobile robots to retain accurate localization system. The paper introduces a stereo vision-based Graph-Simultaneous Localization and Mapping (Graph-SLAM) pipeline custom-tailored to the specificities of vineyard fields. Graph-SLAM is reinforced with a Loop Closure Detection (LCD) based on semantic segmentation of the vine trees. The Mask R-CNN network is applied to segment the trunk regions of images, on which unique visual features are extracted. These features are used to populate the bag of visual words (BoVW s) retained on the formulated graph. A nearest neighbor search is applied to each query trunk-image to associate each unique feature descriptor with the corresponding node in the graph using a voting procedure. We apply a probabilistic method to select the most suitable loop closing pair and, upon an LCD appearance, the 3D points of the trunks are employed to estimate the loop closure constraint to the graph. The traceable features on trunk segments drastically reduce the number of retained BoVWs, which in turn expedites significantly the loop closure and graph optimization, rendering our method suitable for large scale mapping in vineyards. The pipeline has been evaluated on several data sequences gathered from real vineyards, in different seasons, when the appearance of vine trees vary significantly, and exhibited robust mapping in long distances.

This work presents a comparative analysis of human-crafted and automated feature extraction approaches for the discrimination of hand-based activities among eating, drinking and smoking. In this scheme, accelerometer and gyroscope sensors were utilised to capture activity signals. For this reason, wearable devices that embed the aforementioned sensors were employed to collect activity data from 12 office workers. The two approaches that were developed for feature mapping were evaluated equally on the collected dataset. Both the proposed schemes achieved to classify the hand-based activities. However, based on the experimental process, this study shows that the human-crafted features that extracted valuable information from the time and frequency domain of the raw signal measurements outperformed the automated feature mapping that utilised deep learning advances. The relevant results prove that the human-crafted features can recognise hand-based activities with 0.9109 and, on the other hand, automated features with a 0.907 F1 weighted score over the dataset.

Automated guided vehicles operation in human populated factory environments is a challenging task, especially when there is a demand to operate without following fixed paths defined by guide wires, magnetic tape, magnets, or transponders embedded in the floor. The paper at hand introduces a vision-based method enabling safe and autonomous operation of pallet moving vehicles that accommodate pallet detection, pose estimation, docking control and pallet pick up in such industrial environments. A dedicated perception topology relying on monocular vision and laser-based measurements has been applied and installed on-board a novel robotic pallet truck. Pallet detection and pose estimation are performed in two steps. Firstly, a deep neural network is used for the fast isolation of pallets' regions of interest and, secondly, model-based geometrical pattern matching on point cloud data is applied to extract the pallet pose. Robot alignment with candidate pallet is performed with a dedicated visual servoing controller. The developed method has been extensively evaluated both in simulated and real industrial environments with the pallet truck and proved to have real-time performance achieving increased accuracy in navigation, pallet detection and pick-up.

Open Educational Resources (OERs) are often scattered among various sources and may follow different metadata schemata. In addition, they may not include exhaustive annotations; even worse, their subject characterization, if any, may be represented by arbitrary, ad-hoc keywords instead of standard, controlled vocabularies, a fact that stretches up the search space and hampers interoperability. To address this issue, in this paper we propose a twofold method based on two seemingly disjoint technology stacks: machine learning and the semantic web. First, OERs harvested from various repositories are assigned subject terms from a formal, standard thesaurus for a domain of interest, by discovering the semantic matches of the harvesting keyword within the thesaurus ontology. Then, we use word embeddings to represent an item's metadata and compute its similarity with the thesaurus keywords. These word embeddings are learned by a doc2vec model that has been trained with already annotated corpora from the biomedical domain. By combining both worlds, we show that it is possible to produce a reasonable set of thematic suggestions which exceed a certain similarity threshold.

Prediction and modelling of the financial indices is a very challenging and demanding problem because its dynamic, noisy and multivariate nature. Modern approaches have also to challenge the fact that they are dependencies between different global financial indices. All this complexity in combination with the large volume of historic financial data raised the need for advanced machine learning solutions to the problem. This article proposes a Deep Learning approach utilizing Long Short-Term Memory (LSTM) Networks for the modelling and trading of financial indices. The technique is evaluated in the use case of the Athens SE FTSE/ASE Large Cap Index in comparison with a hybrid approach combining Genetic Algorithms and Support Vector Machines with promising results.

Cloud computing involves a set of new technologies that permit the on-demand utilization of computational resources. This is extremely useful, especially when performing high demanding analysis tasks. The flexibility and scalability on providing computational resources has set cloud computing as an emerging technology for data analysis, where problems demand parallelization and are of high computational workload. In this article, a novel Credit-based algorithm targets at providing an efficient and simple (easy to implement) solution to computational resource management challenge by addressing the virtual machine scheduling problem of serving continuously submitted and different analysis tasks in private cloud infrastructure.

MicroRNAs (miRNAs) are small non-coding RNAs, which play a significant role in gene regulation. Predicting miRNA genes is a challenging bioinformatics problem and existing experimental and computational methods fail to deal with it effectively. We developed YamiPred, an embedded classification method that combines the efficiency and robustness of support vector machines (SVM) with genetic algorithms (GA) for feature selection and parameters optimization. YamiPred was tested in a new and realistic human dataset and was compared with state-of-the-art computational intelligence approaches and the prevalent SVM-based tools for miRNA prediction. Experimental results indicate that YamiPred outperforms existing approaches in terms of accuracy and of geometric mean of sensitivity and specificity. The embedded feature selection component selects a compact feature subset that contributes to the performance optimization. Further experimentation with this minimal feature subset has achieved very high classification performance and revealed the minimum number of samples required for developing a robust predictor. YamiPred also confirmed the important role of commonly used features such as entropy and enthalpy, and uncovered the significance of newly introduced features, such as %A-U aggregate nucleotide frequency and positional entropy. The best model trained on human data has successfully predicted pre-miRNAs to other organisms including the category of viruses.

In this work, a novel, multi-space, real-time and robust human tracking system is going to be presented. The system exploits a multi-camera network monitoring the multi-space dynamic environment under interest, detecting and tracking the humans in it. The system is able to handle the dynamic changes of the environment, as well as partial occlusions utilizing virtual top cameras. Furthermore, the system is able to real-time visualize the detection and tracking results on the architectural map of the dynamic environment, as well as a variety of statistics. The visual spatio-temporal analysis of the tracked data are presented in a consolidated form for the overall monitoring area and analytically for each space separately and for each tracked human. These statistics could be also combined with the energy consumption in the area, as well as with other environmental data providing semantic information such as comfort. The overall system is equipped with a number of visual interactive tools providing real-time spatio-temporal human presence analysis offering to the user the opportunity to capture and isolate the areas/spaces with high human presence, the days and times of high human presence, to correlate this information with the potential energy consumption and indicators such as comfort.

MOTIVATION: Single nucleotide polymorphisms (SNPs) are considered the most frequently occurring DNA sequence variations. Several computational methods have been proposed for the classification of missense SNPs to neutral and disease associated. However, existing computational approaches fail to select relevant features by choosing them arbitrarily without sufficient documentation. Moreover, they are limited to the problem of missing values, imbalance between the learning datasets and most of them do not support their predictions with confidence scores. RESULTS: To overcome these limitations, a novel ensemble computational methodology is proposed. EnsembleGASVR facilitates a two-step algorithm, which in its first step applies a novel evolutionary embedded algorithm to locate close to optimal Support Vector Regression models. In its second step, these models are combined to extract a universal predictor, which is less prone to overfitting issues, systematizes the rebalancing of the learning sets and uses an internal approach for solving the missing values problem without loss of information. Confidence scores support all the predictions and the model becomes tunable by modifying the classification thresholds. An extensive study was performed for collecting the most relevant features for the problem of classifying SNPs, and a superset of 88 features was constructed. Experimental results show that the proposed framework outperforms well-known algorithms in terms of classification performance in the examined datasets. Finally, the proposed algorithmic framework was able to uncover the significant role of certain features such as the solvent accessibility feature, and the top-scored predictions were further validated by linking them with disease phenotypes. AVAILABILITY AND IMPLEMENTATION: Datasets and codes are freely available on the Web at http://prlab.ceid.upatras.gr/EnsembleGASVR/dataset-codes.zip. All the required information about the article is available through http://prlab.ceid.upatras.gr/EnsembleGASVR/site.html.

BACKGROUND: During the last years, high throughput experimental methods have been developed which generate large datasets of protein - protein interactions (PPIs). However, due to the experimental methodologies these datasets contain errors mainly in terms of false positive data sets and reducing therefore the quality of any derived information. Typically these datasets can be modeled as graphs, where vertices represent proteins and edges the pairwise PPIs, making it easy to apply automated clustering methods to detect protein complexes or other biological significant functional groupings. METHODS: In this paper, a clustering tool, called GIBA (named by the first characters of its developers' nicknames), is presented. GIBA implements a two step procedure to a given dataset of protein-protein interaction data. First, a clustering algorithm is applied to the interaction data, which is then followed by a filtering step to generate the final candidate list of predicted complexes. RESULTS: The efficiency of GIBA is demonstrated through the analysis of 6 different yeast protein interaction datasets in comparison to four other available algorithms. We compared the results of the different methods by applying five different performance measurement metrices. Moreover, the parameters of the methods that constitute the filter have been checked on how they affect the final results. CONCLUSION: GIBA is an effective and easy to use tool for the detection of protein complexes out of experimentally measured protein - protein interaction networks. The results show that GIBA has superior prediction accuracy than previously published methods.

With the recent high-throughput methods, large datasets of experimentally detected pairwise protein-protein interactions are generated. However, these data suffer from noise, reducing the quality of the information they bring (identification of protein complexes). This paper introduces a novel methodology for detecting protein complexes in a protein-protein interaction graph. Our method initially uses the Markov clustering algorithm and then filters the derived results in order to obtain the best set of clusters that represent protein complexes. The efficiency of our method is shown in experimental results derived from 7 different yeast protein interaction datasets. Moreover, comparisons with 4 other algorithms are performed proving that our method predicts known protein complexes, recorded in the MIPS database, more accurately.

The great heterogeneity of Web based learning systems storing and providing digital e-learning data requires the introduction of interoperability aspects in order to resolve integration problems in a flexible and dynamic way. Learning objects developed and stored in many different places on the Web have a tremendous potential to benefit e-learning in particular and education in general. So the distributed e-learning data integration is significant for the enforcement of novel searching mechanisms in distributed e-learning repositories. Our approach introduces an advanced search mechanism which initializes a semantic model representing the digital content stored in distributed e-learning servers, through the use of ontologies. In order to retrieve useful content to be added to the learning environment while bypassing compatibility problems between different ontological representations of the same domain, we use a mixed approach constituting both keywords and ontologies. Searching tasks are carried out in the metadata level, where information concerning Web digital content is published, managed and stored in the form of a scalable description of knowledge domains

In this paper, an adaptive technique is presented for processing the output of a sensor array, which simultaneously estimates the number of sources and their directions of arrival. The method is based on the reformulation of the problem in the time domain, and the use of the adaptive multi-model partitioning algorithm (MMPA). The adaptive algorithm identifies the dimensionality of the problem (number of sources) using a bank of extended Kalman filters (EKF). The method has the ability of successfully tracking changes in the model structure in real time. This means that, for example, variations in the number of emitting sources are successfully detected. Simulation results demonstrate the performance of the proposed method in multiple source detection and DOA estimation.

Instance selection and feature selection are two orthogonal methods for reducing the amount and complexity of data. Feature selection aims at the reduction of redundant features in a dataset whereas instance selection aims at the reduction of the number of instances. So far, these two methods have mostly been considered in isolation. In this paper, we present a new algorithm, which we call FIS (Feature and Instance Selection) that targets both problems simultaneously in the context of text classificationOur experiments on the Reuters and 20-Newsgroups datasets show that FIS considerably reduces both the number of features and the number of instances. The accuracy of a range of classifiers including Naïve Bayes, TAN and LB considerably improves when using the FIS preprocessed datasets, matching and exceeding that of Support Vector Machines, which is currently considered to be one of the best text classification methods. In all cases the results are much better compared to Mutual Information based feature selection. The training and classification speed of all classifiers is also greatly improved.

Success on the global market has made the adoption of modern management and administration structures of prime importance. The administration of heterogeneous sets of services and applications necessitates the use of information technology, which facilitates cross-organisational process planning and tracking. We describe a virtual private network architecture that uses special monitoring and co-ordination processes in order to achieve efficient delivery of services and activities.

A new emerging area of research and technological development, set forward by the requirements of today's progressive industries, is the paradigm of virtual organisations. A meta-data repository is the logical place for uniformly retaining and managing corporate knowledge within or across different members of such organisations. Many meta-data architectures are not optimized for information retrieval purposes. Taking into consideration the existence of different data types and attributes, many problems arise when we try to successfully merge search results from heterogeneous sources. Motivated by these observations, this paper investigates a new architecture that can support multiple meta-data abstractions and that is readily extensible. Modularity and extensibility are the design emphases.

Nave Bayes (NB) classifier has long been considered a core methodology in text classification mainly due to its simplicity and computational efficiency. There is an increasing need however for methods that can achieve higher classification accuracy while maintaining the ability to process large document collections. In this paper we examine text categorization methods from a perspective that considers the tradeoff between accuracy and scalability to large data sets and large feature sizes. We start from the observation that Support Vector Machines, one of the best text categorization methods cannot scale up to handle the large document collections involved in many real word problems. We then consider bayesian extensions to NB that achieve higher accuracy by relaxing its strong independence assumptions. Our experimental results show that LB, an association-based lazy classifier can achieve a good tradeoff between high classification accuracy and scalability to large document collections...

In the age of information overload intelligent agents have proven themselves as a very useful tool for discovering information of interest on the Web. The information seeking process may be either static, by utilizing existing search engines or using collaborative techniques, or dynamic, by actively browsing the Web. In the second case, agents need to evaluate encountered hyperlinks and choose the promising ones for continuing their autonomous navigation. In this paper we describe a new learning method for identifying interesting links in autonomous information discovery and we present the preliminary results from applying the new method into difficult query domains.