Search

Input-level domain adaptation reduces the burden of a neural encoder without supervision by reducing the domain gap at the input level. Input-level domain adaptation is widely employed in 2D visual domain, e.g., images and videos, but is not utilized for 3D point clouds. We propose the use of input-level domain adaptation for 3D point clouds, namely, point-level domain adaptation. Specifically, we propose to learn a transformation of 3D point clouds by searching the best combination of operations on point clouds that transfer data from the source domain to the target domain while maintaining the classification label without supervision of the target label. We decompose the learning objective into two terms, resembling domain shift and preserving label information. On the PointDA-10 benchmark dataset, our method outperforms state-of-the-art, unsupervised, point cloud domain adaptation methods by large margins (up to + 3.97 % in average).

Feature selection is a technique commonly used in Data Mining and Machine Learning. Traditional feature selection methods, when applied to large datasets, generate a large number of feature subsets. Selecting optimal features within this high dimensional data space is time-consuming and negatively affects the system's performance. This paper proposes a new binary Salp Swarm Algorithm (bSSA) for selecting the best feature set from transformed datasets. The proposed feature selection method first transforms the original data-set using Principal Component Analysis (PCA) and fast Independent Component Analysis (fastICA) based hybrid data transformation methods; next, a binary Salp Swarm optimizer is used for finding the best features. The proposed feature selection approach improves accuracy and eliminates the selection of irrelevant features. We validate our technique on fifteen different benchmark data sets. We conduct an extensive study to measure the performance and feature selection accuracy of the proposed technique. The proposed bSSA is compared to Binary Genetic Algorithm (bGA), Binary Binomial Cuckoo Search (bBCS), Binary Grey Wolf Optimizer (bGWO), Binary Competitive Swarm Optimizer (bCSO), and Binary Crow Search Algorithm (bCSA). The proposed method attains a mean accuracy of 95.26% with 7.78% features on PCA-fastICA transformed datasets. The results show that bSSA outperforms the existing methods for the majority of the performance measures.

Unstructured text contains valuable information for a range of enterprise applications and informed decision making. Text analytics is used to extract valuable insights from unstructured big data. Among the most significant challenges of text analytics, quality and usability are critical in affecting the outcome of the analytical process. The enhancement in usability is important for the exploitation of unstructured data. Most of the existing literature focuses on the usability of structured data as compared to unstructured data whereas big data usability has been discussed merely in the context of its assessment. The existing approaches do not provide proper guidelines on the usability enhancement of unstructured data. In this study, a rigorous systematic literature review using PRISMA framework has been conducted to develop a model enhancing the usability of unstructured data bridging the research gap. The recent approaches and solutions for text analytics have been investigated thoroughly. The usability issues of unstructured text data and their consequences on data preparation for analytics have been identified. Defining the usability dimensions for unstructured big data, identification of the usability determinants, and developing a relationship between usability dimension and determinants to derive usability rules are the significant contributions of this research and are integrated to formulate the usability enhancement model. The proposed model is the major outcome of the research. It contributes to make unstructured data usable and facilitates the data preparation activities with more valuable data that eventually improve the analytical process.

Development of new methods and tools for formalization and representation of complex knowledge in the context of the creation of intelligent systems remains in the scope of scientific research. Modern trends aim to automate the knowledge formalization and representation by using various sources of information, in particular, spreadsheets. This paper proposes a novel approach to the semi-automatic formalization and representation of the engineering knowledge in the form of conceptual models and knowledge base codes from spreadsheet data. Our approach comprises three phases: (I) rule-driven data transformation source spreadsheet tables to a specific canonical form (data preprocessing), (II) domain knowledge formalization and representation via the extraction and aggregation of conceptual model fragments from canonicalized tables, (III) model-driven synthesizing knowledge base and source codes from a domain model. The approach is implemented by our tools: TABBYXL provides the development of a software application for the spreadsheet data canonicalization; Personal Knowledge Base Designer is used to build and aggregate conceptual models fragments, as well as to construct a target knowledge base and to generate source codes. Our case study on the industrial safety inspection (ISI) demonstrates that the approach is fully suitable for prototyping knowledge bases containing decision-making rules.

Demand side management (DSM) programs are an integral part of the modern grid. Most of these DSM programs are designed to work at household and hour level. The optimization problems in these DSM programs are guided by the forecasted load. An error in the hour ahead load forecasting results in a suboptimal solution entailing economic cost to both the utility and the customers. Predicting loads at a fine granularity (e.g., households) is challenging due to a large number of (known or unknown) factors affecting power consumption. At larger scales (e.g., clusters of consumers), since the inherent stochasticity and fluctuations are averaged out, the problem becomes substantially easier. Many techniques have been proposed to predict loads of clusters of consumers in various localities with great accuracy. Also, these techniques generally utilize sociological and weather information and work better on data from a particular locality. In this paper, a new technique called Past Vector Similarity (PVS) has been proposed to predict electricity load one hour ahead at the level of an individual household. The proposed strategy is based on load information only and does not require calendar, weather or any other attributes. In fact, the idea is to extract the exact load patterns of individual households corresponding to their routine and socio-economic values. Consequently, the technique makes use of the recent past vector and generate similar patterns for the prediction of future load profiles. Furthermore, the ensemble of these similar loads is an efficient prediction of electricity. PVS has just two parameters due to which it can be applied to the smaller dataset without overfitting issue. Moreover, due to the parallel nature of PVS, it can be scaled for a large number of customers without computation burden. The proposed PVS has been assessed empirically for two distinct datasets from Australia and Sweden. The simulation results demonstrate that the PVS significantly outperforms other state-of-the-art forecasting methods in terms of accuracy.

RDF and Property Graphs are data models that are being used to represent Knowledge Graphs. The definition of methods to transform RDF data into Property graph data is fundamental to allow interoperability among the systems using these models. Although both models are based on a graph structure, they have special features that complicate the definition of data transformation methods. This article presents an ontology-based approach to transform (automatically) property graphs into RDF graphs. The ontology, called PGO, defines a set of terms that allows describing the elements of a property graph. The algorithm corresponding to the transformation method is described, and some properties of the method are discussed (complexity, data preservation, and monotonicity). The results of an experimental evaluation are also presented.

The traditional big-data analytical approaches use data clustering as small buckets while providing distributed computation among different child nodes. These approaches bring the issues especially concerning network capacity, specialized tools and applications not capable of being trained in a short period. Furthermore, raw data generated through IoT forming big data comes with the capability of producing highly unstructured and heterogeneous form of data. Such form of data grows into challenging task for the real-time analytics. It is highly valuable to have computational values available locally instead of through distributed resources to reduce real-time analytical challenges. This paper proposes a fusion of three different data models like relational, semantical, and big data based data and metadata involving their issues and enhanced capabilities. A case study is used to represent data fusion in action from RDB to Resource Description Framework. Whereas, issues and their feasible solutions are also being discussed in this paper.

Data leakage is a growing insider threat in information security among organizations and individuals. A series of methods has been developed to address the problem of data leakage prevention (DLP). However, large amounts of unstructured data need to be tested in the big data era. As the volume of data grows dramatically and the forms of data become much complicated, it is a new challenge for DLP to deal with large amounts of transformed data. We propose an adaptive weighted graph walk model to solve this problem by mapping it to the dimension of weighted graphs. Our approach solves this problem in three steps. First, the adaptive weighted graphs are built to quantify the sensitivity of the tested data based on its context. Then, the improved label propagation is used to enhance the scalability for fresh data. Finally, a low-complexity score walk algorithm is proposed to determine the ultimate sensitivity. Experimental results show that the proposed method can detect leaks of transformed or fresh data fast and efficiently.