Pruning Deep Reinforcement Learning for Dual User Experience and Storage Lifetime Improvement on Mobile Devices

Author(s):  
Chao Wu ◽  
Yufei Cui ◽  
Cheng Ji ◽  
Tei-Wei Kuo ◽  
Chun Jason Xue
Author(s):  
Weixiang Xu ◽  
Xiangyu He ◽  
Tianli Zhao ◽  
Qinghao Hu ◽  
Peisong Wang ◽  
...  

Large neural networks are difficult to deploy on mobile devices because of intensive computation and storage. To alleviate it, we study ternarization, a balance between efficiency and accuracy that quantizes both weights and activations into ternary values. In previous ternarized neural networks, a hard threshold Δ is introduced to determine quantization intervals. Although the selection of Δ greatly affects the training results, previous works estimate Δ via an approximation or treat it as a hyper-parameter, which is suboptimal. In this paper, we present the Soft Threshold Ternary Networks (STTN), which enables the model to automatically determine quantization intervals instead of depending on a hard threshold. Concretely, we replace the original ternary kernel with the addition of two binary kernels at training time, where ternary values are determined by the combination of two corresponding binary values. At inference time, we add up the two binary kernels to obtain a single ternary kernel. Our method dramatically outperforms current state-of-the-arts, lowering the performance gap between full-precision networks and extreme low bit networks. Experiments on ImageNet with AlexNet (Top-1 55.6%), ResNet-18 (Top-1 66.2%) achieves new state-of-the-art.


Author(s):  
Ihssan Alkadi

There are many steps involved with securing a cloud system and its applications (SaaS) and developed ones in (PaaS). Security and privacy issues represent the biggest concerns to moving services to external clouds (Public). With cloud computing, data are stored and delivered across the Internet. The owner of the data does not have control or even know where their data are being stored. Additionally, in a multi-tenant environment, it may be very difficult for a cloud service provider to provide the level of isolation and associated guarantees that are possible with an environment dedicated to a single customer. Unfortunately, to develop a security algorithm that outlines and maps out the enforcement of a security policy and procedure can be a daunting task. A good security algorithm presents a strategy to counter the vulnerabilities in a cloud system. This chapter covers the complete overview, comparative analysis of security methods in Cloud Applications in STEM Education and the introduction of a new methodology that will enforce cloud computing security against breaches and intrusions. Much light will be shed on existing methodologies of security on servers used for cloud applications in STEM education and storage of data, and several methods will be presented in addition to the newly developed method of security in cloud-based servers, such as the MIST (Alkadi). Not only can cloud networks be used to gather sensitive information on multiple platforms, also there are needs to prevent common attacks through weak password recovery, retrieval, authentication, and hardening systems; otherwise hackers will spread cyber mayhem. Discussion of current security issues and algorithms in a real world will be presented. Different technologies are being created and in constant competition to meet the demands of users who are generally “busy”. The selling point of these technologies is the ability to address these demands without adding more to any workloads. One of the demands often discussed is that users want to have their digital information accessible from anywhere at any time. This information includes documents, audio libraries, and more. Users also demand the ability to manage, edit and update this information regardless of physical location. Somewhat recently, mobile devices such as laptops, tablets, and smartphones have provided these abilities. This is no small feat as vendors and providers have reduced the size of these devices to increase mobility. However, as the amount of personal information that users are wanting to access has grown exponentially, manipulation and storage of it require more capable devices. To meet increased demands, increasing the capabilities of mobile devices may be impractical. Making mobile devices more powerful without technological advancement would require that the device be larger and use more resources such as battery life and processing power to function properly. Storing all of a user's information on a mobile device that travels everywhere also adds vulnerability risks. The best technical solution to having a user's information accessible is some sort of online storage where there is the convenience to store, manipulate and retrieve data. This is one of the most practical applications for the concept of cloud computing in STEM education. As storage capabilities and Internet bandwidth has increased, so has the amount of personal data that users store online. And today, the average user has billions of bytes of data online. Access is everywhere and whenever is needed. As everyone started doing so, people want their data safe and secure to maintain their privacy. As the user base grew in size, the number of security issues of the personal data started to become increasingly important. As soon as someone's data are in the remote server, unwanted users or “hackers” can have many opportunities to compromise the data. As the online server needs to be up and running all the time, the only way to secure the cloud server is by using better passwords by every user. By the same token, the flaws in the password authentication and protection system can also help unwanted users to get their way to other people's personal data. Thus, the password authentication system should also be free from any loopholes and vulnerabilities.


Author(s):  
Franck Amadieu ◽  
Charly Pecoste ◽  
Claudette Mariné ◽  
Cécile van de Leemput ◽  
Colin Lescarret

This chapter addresses the issue of tablets acceptance for studying. An experiment was carried out to test the effects of specific studying tasks experienced by students with no previous experiences with tablets on the perceived usability and usefulness of tablets. Students had to perform a high-compatible task (i.e. navigation/reading task) and a low-compatible task (i.e. writing task) for tablets. Subjective measures of usability, usefulness and use intention were designed to be more specific to the type of task than the classical measures used in the Technology Acceptance Model approach (Davis, 1989). Participants rated their answers before and after performing the tasks with a tablet. The results showed that the perceived usability and usefulness of tablets increased after the high-compatible task while their decreased with the low-compatible task. The findings stressed the need to consider the real user experience and to use more task-oriented measures to investigate the acceptance of mobile devices for studying.


2015 ◽  
Vol 6 (1) ◽  
pp. 34-43
Author(s):  
Mark Bruce Freeman

There has been a dramatic shift in the interaction methods of mobile devices over the past decade. From devices simply being able to make phone calls to being able to handle complex tasks traditionally performed on personal computers (PCs); this change has led to new interaction issues that need to be understood during the application development process, particularly as these devices now commonly incorporate a touch-screen as their primary source of input. Currently, the methods of conducting software user experience testing of these devices employs techniques that were developed for PCs, however mobile devices are used within different contexts of use. This research initially reviews the current methods for user experience testing of applications running on mobile devices and then presents, through a proof-of-concept approach, an innovative method for conducting user experience testing employing actual devices.


2020 ◽  
Vol 12 (2) ◽  
pp. 1-21
Author(s):  
Helge Nissen ◽  
Monique Janneck

Participants increasingly use mobile devices, especially smartphones, to fill out online questionnaires. However, standard questionnaire templates are often not optimized for presentation on smartphones, raising the question of whether an unfavorable layout may influence the survey results. In this study, interaction with questionnaires on different devices was investigated regarding processing time, data quality, and user experience of the questionnaire itself. Several standard and newly developed questionnaire layout templates were evaluated by means of an online study (N=301). Results show that processing times are higher on smartphones compared to desktop computers. However, there were no differences regarding data quality. The comparison of different mobile layouts among smartphone users revealed effects on processing time and user experience. Design recommendations are derived.


2020 ◽  
pp. 697-720
Author(s):  
Franck Amadieu ◽  
Charly Pecoste ◽  
Claudette Mariné ◽  
Cécile van de Leemput ◽  
Colin Lescarret

This chapter addresses the issue of tablets acceptance for studying. An experiment was carried out to test the effects of specific studying tasks experienced by students with no previous experiences with tablets on the perceived usability and usefulness of tablets. Students had to perform a high-compatible task (i.e. navigation/reading task) and a low-compatible task (i.e. writing task) for tablets. Subjective measures of usability, usefulness and use intention were designed to be more specific to the type of task than the classical measures used in the Technology Acceptance Model approach (Davis, 1989). Participants rated their answers before and after performing the tasks with a tablet. The results showed that the perceived usability and usefulness of tablets increased after the high-compatible task while their decreased with the low-compatible task. The findings stressed the need to consider the real user experience and to use more task-oriented measures to investigate the acceptance of mobile devices for studying.


2019 ◽  
Vol 2019 ◽  
pp. 1-18 ◽  
Author(s):  
Paulo A. L. Rego ◽  
Fernando A. M. Trinta ◽  
Masum Z. Hasan ◽  
Jose N. de Souza

Mobile cloud computing is an approach for mobile devices with processing and storage limitations to take advantage of remote resources that assist in performing computationally intensive or data-intensive tasks. The migration of tasks or data is commonly referred to as offloading, and its proper use can bring benefits such as performance improvement or reduced power consumption on mobile devices. In this paper, we face three challenges for any offloading solution: the decision of when and where to perform offloading, the decision of which metrics must be monitored by the offloading system, and the support for user’s mobility in a hybrid environment composed of cloudlets and public cloud instances. We introduce novel approaches based on machine learning and software-defined networking techniques for handling these challenges. In addition, we present details of our offloading system and the experiments conducted to assess the proposed approaches.


Sign in / Sign up

Export Citation Format

Share Document