In machine learning, model compression is vital for resource-constrained Internet of Things (IoT) devices, such as unmanned aerial vehicles (UAVs) and smart phones. Currently there are some state-of-the-art (SOTA) compression methods, but little study is conducted to evaluate these techniques across different models and datasets. In this paper, we present an in-depth study on two SOTA model compression methods, pruning and quantization. We apply these methods on AlexNet, ResNet18, VGG16BN and VGG19BN, with three well known datasets, Fashion-MNIST, CIFAR-10, and UCI-HAR. Through our study, we draw the conclusion that, applying pruning and retraining could keep the performance (average less than 0.5% degrade) while reducing the model size (at 10× compression rate) on spatial domain datasets (e.g. pictures); the performance on temporal domain datasets (e.g. motion sensors data) degrades more (average about 5.0% degrade); the performance of quantization is related with the pruning rate and the network architecture. We also compare different clustering methods and reveal the impact on model accuracy and quantization ratio. Finally, we provide some interesting directions for future research.
With the rapid development of today’s society, the traditional social and government management model has gradually emerged drawbacks, modern social management requires multi-sectoral integration of government resources to provide joint management and services. In addition, because the speed of economic development in the new era is far ahead of social management, the adjustment of the pattern of interests in social management and more weak links, resulting in conflicts and occasional incidents in the community increased, which requires social management work to reach a higher level. In recent years, China has actively explored and innovated in social governance and continuously improved the level of social governance. But the overall effect and management level still need to be further improved. This paper first expounds the basic connotation and category of social governance, and then comprehensively analyzes the problems existing in social governance. Combined with modeling and simulation system, intelligent governance support, GIS [9, 10] and other technologies, after combing the business, the social comprehensive management information system is designed. The main application modules of the system include provincial and municipal standard platform module, provincial supervision module, data docking standard module, system docking module.
5G networks have introduced new technologies and paradigms to support new use cases with very demanding quality of service (QoS) requirements, in terms of metrics such as delay or reliability. Accordingly, operators need new tools to measure these metrics in different realistic, often extreme, conditions, so that they can evaluate the degree of fulfilment of their service levels. In this work we propose a simple practical framework to evaluate the delay between two nodes in such a cellular network. This framework allows evaluating the delay in the uplink and downlink channels independently. We have validated the proposed framework on top of a real 5G network, by measuring the delay as a function of the requested data rates of the network in different network slices.
This study proposes a conference bridge that cooperates with both the WebRTC layered simulcast and the Software Defined Networking architecture in order to improve WebRTC video streaming. The proposed bridge divides the functionality of a classic Selective Forwarding Unit into two parts. The selection of layers is performed by the SDN controller and the forwarding of layered video is still accomplished by the bridge. The bridge and the SDN controller exchange data on the state of the transmitted video stream and the state of the network. The proposed solution was implemented in the Jitsi Videobridge and tested in the GEANT testbed network. The results showed that our solution significantly reduces problems related to available throughput overshooting, which is typical for layered simulcast.
Android applications are proliferating, which has led to the rise of android malware. Many research studies have proposed various detection frameworks for android malware detection. Literature suggests that static malware detection techniques are practical and assuring for detecting android malware. This paper presents a thorough survey of data mining-based static malware detection. We briefly discuss the growth of android malware and current detection techniques and offer a comprehensive analysis and summary of studies for each data mining-based malware detection phase, such as data acquisition, preprocessing, feature extraction, learning algorithms, and evaluation. Finally, we highlight some challenges and open issues in data mining-based android malware detection. This review will help understand the complete picture of static android malware detection and serve as a basis for malware detection in general.
