Search

Abstract Activation functions are essential in deep learning, and the rectified linear unit (ReLU) is the most widely used activation function to solve the vanishing gradient problem. However, owing to the dying ReLU problem and bias shift effect, deep learning models using ReLU cannot exploit the potential benefits of negative values. Numerous ReLU variants have been proposed to address this issue. In this study, we propose Dynamic Parametric ReLU (DPReLU), which can dynamically control the overall functional shape of ReLU with four learnable parameters. The parameters of DPReLU are determined by training rather than by humans, thereby making the formulation more suitable and flexible for each model and dataset. Furthermore, we propose an appropriate and robust weight initialization method for DPReLU. To evaluate DPReLU and its weight initialization method, we performed two experiments on various image datasets: one using an autoencoder for image generation and the other using the ResNet50 for image classification. The results show that DPReLU and our weight initialization method provide faster convergence and better accuracy than the original ReLU and the previous ReLU variants.

In the recent era of information explosion, exploring event from social networks has recently been a crucial task for many applications. To derive valuable comprehensive and thorough insights on social events, visual analytics (VA) system have been broadly used as a promising solution. However, due to the enormous social data volume with highly diversity and complexity, the number of event exploration tasks which can be enabled in a conventional real-time visual analytics systems has been limited. In this article, we introduce SocioPedia+, a real-time visual analytics system for social event exploration in time and space domains. By introducing the dimension of social knowledge graph analysis into the system multivariate analysis, the process of event explorations in SocioPedia+ can be significantly enhanced and thus enabling system capability on performing full required tasks of visual analytics and social event explorations. Furthermore, SocioPedia+ has been optimized for visualizing event analysis on different levels from macroscopic (events level) to microscopic (knowledge level). The system is then implemented and investigated with a detailed case study for evaluating its usefulness and visualization effectiveness for the application of event explorations.

This study suggests a method to select core data that will be helpful for machine learning. Specifically, we form a two-dimensional distribution based on the similarity of the training data and compose grids with fixed ratios on the distribution. In each grid, we select data based on the distribution consistency (DC) of the target class data and examine how it affects the classifier. We use CIFAR-10 for the experiment and set various grid ratios from 0.5 to 0.005. The influences of these variables were analyzed with the use of different training data sizes selected based on high-DC, low-DC (inverse of high DC), and random (no criteria) selections. As a result, the average point accuracy at 0.95% (±0.65) and the point accuracy at 1.54% (±0.59) improved for the grid configurations of 0.008 and 0.005, respectively. These outcomes justify an improved performance compared with that of the existing approach (data distribution search). In this study, we confirmed that the learning performance improved when the training data were selected for very small grid and high-DC settings.

This paper presents a cross-cultural crowdsourcing platform, called OurPlaces, where people from different cultures can share their spatial experiences. We built a three-layered architecture composed of: (i) places (locations where people have visited); (ii) cognition (how people have experienced these places); and (iii) users (those who have visited these places). Notably, cognition is represented as a paring of two similar places from different cultures (e.g., Versailles and Gyeongbokgung in France and Korea, respectively). As a case study, we applied the OurPlaces platform to a cross-cultural tourism recommendation system and conducted a simulation using a dataset collected from TripAdvisor. The tourist places were classified into four types (i.e., hotels, restaurants, shopping malls, and attractions). In addition, user feedback (e.g., ratings, rankings, and reviews) from various nationalities (assumed to be equivalent to cultures) was exploited to measure the similarities between tourism places and to generate a cognition layer on the platform. To demonstrate the effectiveness of the OurPlaces-based system, we compared it with a Pearson correlation-based system as a baseline. The experimental results show that the proposed system outperforms the baseline by 2.5% and 4.1% in the best case in terms of MAE and RMSE, respectively.

The acceptance and usability of context-aware systems have given them the edge of wide use in various domains and has also attracted the attention of researchers in the area of context-aware computing. Making user context information available to such systems is the center of attention. However, there is very little emphasis given to the process of context representation and context fusion which are integral parts of context-aware systems. Context representation and fusion facilitate in recognizing the dependency/relationship of one data source on another to extract a better understanding of user context. The problem is more critical when data is emerging from heterogeneous sources of diverse nature like sensors, user profiles, and social interactions and also at different timestamps. Both the processes of context representation and fusion are followed in one way or another; however, they are not discussed explicitly for the realization of context-aware systems. In other words most of the context-aware systems underestimate the importance context representation and fusion. This research has explicitly focused on the importance of both the processes of context representation and fusion and has streamlined their existence in the overall architecture of context-aware systems’ design and development. Various applications of context representation and fusion in context-aware systems are also highlighted in this research. A detailed review on both the processes is provided in this research with their applications. Future research directions (challenges) are also highlighted which needs proper attention for the purpose of achieving the goal of realizing context-aware systems.

Many studies have investigated the management of data delivered over sensor networks and attempted to standardize their relations. Sensor data come from numerous tangible and intangible sources, and existing work has focused on the integration and management of the sensor data itself. The data should be interpreted according to the sensor environment and related objects, even though the data type, and even the value, is exactly the same. This means that the sensor data should have semantic connections with all objects, and so a knowledge base that covers all domains should be constructed. In this paper, we suggest a method of domain terminology collection based on Wikipedia category information in order to prepare seed data for such knowledge bases. However, Wikipedia has two weaknesses, namely, loops and unreasonable generalizations in the category structure. To overcome these weaknesses, we utilize a horizontal bootstrapping method for category searches and domain-term collection. Both the category-article and article-link relations defined in Wikipedia are employed as terminology indicators, and we use a new measure to calculate the similarity between categories. By evaluating various aspects of the proposed approach, we show that it outperforms the baseline method, having wider coverage and higher precision. The collected domain terminologies can assist the construction of domain knowledge bases for the semantic interpretation of sensor data.

This paper proposes a multilayered quadrotor control method that can move the quadrotor to the desired goal while resisting disturbance. The proposed control system is modular, convenient to design and verify, and easy to extend. It comprises three layers: a physical layer, a displacement control layer, and an attitude control layer. The displacement control layer considers the movement of the vehicle, while the attitude control layer controls its attitude. The physical layer deals with the physical operation of the vehicle. The two control layers use a mathematical method to provide minute step-by-step control. The proposed control system effectively combines the three layers to achieve drift stabilization.