Binbin Yong

Neural network model with Monte Carlo algorithm for electricity demand forecasting in Queensland

With the rapid growth over the past few decades, people are consuming more and more electrical energies. In order to solve the contradiction between supply and demand to minimize electricity cost, it is necessary and useful to predict the electricity demand. In this paper, we apply an improved neural network algorithm to forecast the electricity, and we test it on a collected electricity demand data set in Queensland to verify its performance. There are two contributions in this paper. Firstly, comparing with backpropagation (BP) neural network, the results show a better performance on this improved neural network. Secondly, the performance on various hidden layers shows that different dimension of hidden layer in this improved neural network has little impact on the Queenslands electricity demand forecasting.

IoT-based intelligent fitness system

Abstract With the global economic growth, fitness club is developing rapidly in the world. Meanwhile, the fitness industry is booming especially for urban white-collar population. In the circumstances, people need more scientific and practical guidance to build their body. In this paper, we design an Internet of things (IoT) based fitness system to monitor the health statuses of exercisers. The system provides guidance for exercisers. When exercising, the exercise data is collected by sensors and fitness band. Subsequently, these data are sent to the system to be analyzed. With the help of artificial intelligence technology, the system can extract useful guidance information for users’ body building. In this paper, we will describe the details of the system and further reach out to the implementation technologies. The design of this kind of system is a trend for the future fitness application.

Intelligent monitor system based on cloud and convolutional neural networks

Nowadays, cloud-based services are widely developed. The deployment of cloud technology has boosted the development and application of web services. It reduces the overhead of software virtual machine, and supports a wider range of operating systems. Moreover, it enhances the utilization of infrastructure. With the development of artificial intelligence (AI) technology, especially artificial neural network (ANN), intelligent monitor systems are being raised and developed in our daily life. However, a simple task with a single ANN costs a lot of time and computation resources. Hence, we propose using a cloud-based system to share computation resources for ANN to reduce redundant computation. In this paper, we present an intelligent monitor system, which is based on cloud technology, to provide intelligent monitor services. The system is designed with hybrid convolutional neural networks. It has been used for several intelligent monitor tasks, such as scene change detection, stranger recognition, facial expression recognition and action recognition.

Parallel GPU-based collision detection of irregular vessel wall for massive particles

In this paper, we present a novel GPU-based limit space decomposition collision detection algorithm (LSDCD) for performing collision detection between a massive number of particles and irregular objects, which is used in the design of the Accelerator Driven Sub-Critical (ADS) system. Test results indicate that, the collisions between ten million particles and the vessel can be detected on a general personal computer in only 0.5 s per frame. With this algorithm, the collision detection of maximum sixty million particles are calculated in 3.488030 s. Experiment results show that our algorithm is promising for fast collision detection.

Anomaly detection in ad-hoc networks based on deep learning model: A plug and play device

Abstract Ad-hoc network is a temporary self-organizing network that needs no fixed infrastructure. So it has been applied extensively in many areas requesting temporary communication such as military field, emergency disaster relief and road traffic. While, due to the feature of self-organization and wireless communication channels, ad-hoc network is more vulnerable to various attacks compared to the traditional network. In this paper, we proposed a plug and play device to detect Denial of Service (DoS) and privacy attacks. This device mainly includes capture tool and deep learning detection model. Capture tool is used to grab packets in ad-hoc networks, deep learning detection model is used for detecting attacks. An alarm will be triggered if the detected result is attack. In this way, we can avoid the detected attack to spreading out in larger scale. The proposed method can be used as the second line of dense to issue the early-warning signal. In the experiment, first, we use Deep neural network (DNN) detection model to detect DoS attacks; next, we use DNN, Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) detection model to detect XSS and SQL attacks. The results show that these detection models can achieve very high Accuracy, Precision, Recall and F 1 − s c o r e . In addition, the time efficiency among the CNN, the LSTM and the DNN is in acceptable range. It proofs that the proposed method can be effectively applied for attack detection. It is important to note that the proposed method can be extended to all other attacks with little modification in ad-hoc networks.

Statistical study of characteristics of online reading behavior networks in university digital library

Composed of a large number of interacting nodes, an online reading behavior network (ORBN) in a university digital library constitutes a large-scale online social network, which can be modeled as a complex system. Online social networks have been investigated intensively by many researchers over the past several years. However, there is little research on the online reading behaviors in university digital libraries. In this paper, we investigate the statistical characteristics of ORBNs in university digital libraries. This study reveals that the degree distribution of an online reading behavior network obeys the exponential distribution. In addition, the small-world phenomenon is observed in ORBNs. This study also compares the statistical characteristics of different types of ORBNs. The results show that the statistical characteristics of online reading behavior sub-networks remain consistent over multiple school years. However, over every four school years, differences are identified between the global ORBN and its sub-networks.

Derivative-based acceleration of general vector machine

General vector machine (GVM) is one of supervised learning machine, which is based on three-layer neural network. It is capable of constructing a learning model with limited amount of data. Generally, it employs Monte Carlo algorithm (MC) to adjust weights of the underlying network. However, GVM is time-consuming at training and is not efficient when compared with other learning algorithm based on gradient descent learning. In this paper, we present a derivative-based Monte Carlo algorithm (DMC) to accelerate the training of GVM. Our experimental results indicate that DMC algorithm is faster than the original MC method. Specifically, the training time of our DMC algorithm in GVM for function fitting is also less than some gradient descent-based methods, in which we compare DMC with back-propagation neural network. Experimental results indicate that our algorithm is promising for training GVM.

GVM Based Intuitive Simulation Web Application for Collision Detection

Abstract Computer simulation, which has been proved to be an effective approach to problem solving, is nowadays widely used in modern science. However, it requires a lot of computing resources, which are difficult for general users to acquire. In this paper, we design a Web based system to implement on-line simulation system for ordinary users. As a useful example, the simulation of one type of collision detection model is presented in this paper. Moreover, the software application of simulation is offered as a service on Web. Meanwhile, the incorporation of general vector machine (GVM, a type of neural network) to intelligently predict the relationship between simulation parameters and computation resources is presented, which could further provide more information for system monitoring and scheduling. The system has demonstrated efficiency and intuitiveness for users of this type of applications.

L4eRTL: a robust and secure real-time architecture with L4 microkernel and para-virtualised PSE51 partitions

Different classes of applications exhibit different demands on todays heterogeneous computing systems whose complexity will incur further challenges in ensuring system safety and application security for certain required functionality of these systems. This paper proposes a component-based robust and secure real-time architecture, L4eRTL, which decomposes complex systems into sub-modules and distributes them to POSIX-enabled partitions through para-virtualisation. L4eRTL is designed and implemented based on low-level L4 microkernel and several essential para-virtualised components by IRQ virtualisation, time virtualisation, clock virtualisation, and memory virtualisation. Micro-benchmark programs and sample applications have been conducted to demonstrate the effectiveness of L4eRTL in supporting multiple isolated PSE51 environments for system safety and application security.

GPU Based Simulation of Collision Detection of Irregular Vessel Walls

Collision detection is a commonly used technique in the fields of computer games, physical simulation , virtual technology, computing and animation. When simulating the process of particle collision of ADS (Accelerator Driven Sub-Critical) system, complex and irregular vessel walls need to be considered. Generally, an irregular vessel wall is a curve surface, which cannot be defined as an exact mathematical function, and it is difficult to calculate the distance between particles and the wall directly. In this paper, we present an algorithm to perform collision detection between particles and irregular wall. When the number of particles reaches the level of 106, our algorithm implements a considerable improvement in performance if running on GPU, nearly 10 times faster than running on CPU. Results have demonstrated that our algorithm is promising.