International Journal of Semantic Computing

Research has shown that visual information of multimedia is critical in highly-skilled applications, such as biomedicine and life sciences, and a certain visual reasoning process is essential for meaningful search in a timely manner. Relevant image characteristics are learned and verified with accumulated experiences during the reasoning processes. However, such processes are highly dynamic and elusive to computationally quantify and therefore challenging to analyze, let alone to make the knowledge sharable across users. In this paper we study real-time human visual reasoning processes with the aid of gaze tracking devices. Temporal and spatial representations are proposed for gaze modeling, and a visual reasoning retrieval system utilizing in-memory computing under Big Data ecosystems is designed for real-time search of similar reasoning models. Simulated data derived from human subject experiments show that the system has a reasonably high accuracy and provides predictive estimations for hardware requirements versus data sizes for exhaustive searches. Comparison between various visual action classifiers show challenges in modeling visual actions. The proposed system provides a theoretical framework and computing platform for advancement in visual semantic computing, as well as potential applications in medicine, social science, and arts.

Fuelled by the advancement in multimedia technologies, users across the world have witnessed the proliferation of online videos. Compared with the visual content of these videos, the textual content, for example, titles, tags, or descriptions, has been more broadly exploited in the real-world video data mining or information retrieval tasks. To enhance the understanding of videos, and improve the performance of the tasks such as automatic video annotation, video clustering, and cross-modal tag cleansing, the textual and visual content of videos are combined, through various methods. However, the absence of an empirical study on the properties of these contents makes them less solid to gain satisfactory performance. Therefore, in this paper, we conduct this study to verify the properties of textual content and draw insights from these analyses to promote further developments in video data mining that combine the two contents.

Organizations have come to realize that storing their databases in the Cloud rather than in-house data centers is cheaper and more flexible. However, companies are still concerned about the privacy and the security of their data. Encrypting the whole database before uploading it to the Cloud solves the security issue. But querying the database requires downloading and decrypting the entire dataset, which is impractical.

This paper proposes a new scheme for protecting the privacy and integrity of spatial data stored in the Cloud while being able to execute range queries efficiently. Data objects are encrypted and sorted using Z-order space-filling curve. An index is built on top of the encrypted data to be utilized by the Service Provider to identify and retrieve a superset of data objects that contains the answers to the query. Many simulation experiments have been carried out to measure the performance of the proposed scheme in terms of the redundancy in data retrieved. The experimental results show that the proposed scheme outperforms the most recent scheme by Kim et al. in terms of data redundancy.

Raising children is challenging and requires lots of care. Parents always have to provide proper care to their children in time, like hydration and clothing. However, it is difficult to always stay alert or be aware of the care required at proper moments. One reason is that parents nowadays are busy. This especially applies to single parent, or the one who needs to raise multiple children.

This paper presents the use of an integrated multi-sensors together with a mobile application to help keep track of unusual situations concerning a child. By monitoring the changes in surrounding temperature, motions, and air pressure acquired from the sensors, our mobile application can infer the physiological needs of the children with the heat equilibrium assumption. As the thermal environment in the human body is mainly governed by the heat balance equation, we fuse all available sensor readings to the equation so as to estimate the change in situation of a child over a certain period of time. Our system can then notify the parents of the necessary care, including hydration, dining, clothing and ear barotrauma relieving. The proposed application can greatly relieve some of the mental load and pressure of the parents in taking care of children.

Visual concept recognition is an active research field in the last decade. Related to this attention, deep learning architectures are showing great promise in various computer vision domains including image classification, object detection, event detection and action recognition in videos. In this study, we investigate various aspects of convolutional neural networks for visual concept recognition. We analyze recent studies and different network architectures both in terms of running time and accuracy. In our proposed visual concept recognition system, we first discuss various important properties of popular convolutional network architecture under consideration. Then we describe our method for feature extraction at different levels of abstraction. We present extensive empirical information along with best practices for big data practitioners. Using these best practices we propose efficient fusion mechanisms both for single and multiple network models. We present state-of-the-art results on benchmark datasets while keeping computational costs at low level. Our results show that these state-of-the-art results can be reached without using extensive data augmentation techniques.

The advent of new web technologies and the explosion of available information online led to an information overload. During this information revolution blogs have become considerably mainstream as a media of providing news. Although there are several arguments about their validity and credibility the large amount of blogs currently available require the usage of advanced techniques for the collection, analysis, mining and efficient querying of the available information. To this direction we present BlogSearch, a novel platform allowing aggregating, indexing and searching blog articles. The information is modelled using a novel RDF/S Ontology named Blogs Ontology and published as Linked Open Data. In addition, two sets of APIs are provided for inserting, updating and searching information whereas the platform provides also graphical user interfaces (GUIs) for searching and inserting information. To the best of our knowledge our platform is the only one currently available publishing blog articles as Linked Open Data and simultaneously providing APIs and GUIs for aggregating, inserting and searching articles.

Deep learning is a branch of machine learning that tries to model high-level abstractions of data using multiple layers of neurons consisting of complex structures or non-liner transformations. With the increase of the amount of data and the power of computation, neural networks with more complex structures have attracted widespread attention and been applied to various fields. This paper provides an overview of deep learning in neural networks including popular architecture models and training algorithms.