Mikhail Zhelezin1,2, Stanislav Pavlov1,2,3, PhD Valeria Lakshina1,2, Dr. Prof. Yury Fedosenko3, 1Huawei Research Center, Nizhny Novgorod, Russia, 2National Research University Higher School of Economics" Nizhny Novgorod, Russia, 3Volga State University of Water Transport, Nizhny Novgorod, Russia
We are conducting research aimed at accelerating neural networks on a CPU. This acceleration will be associated with the solution of the computational graph planning problem. In the modern world, large neural networks are computed on accelerators (GPU, NPU, TPU), since their architecture is designed for numerous calculations of matrix multiplications and vector operations. Many modern approaches do not take into account factors that can speed up the inference of neural networks, such as the number of cores, cache size, bus bandwidth, etc. Also, many researchers underestimate simple heuristics that can significantly speed up the computation of neural networks. In this article, we show how simple heuristics can efficiently solve the scheduling problem and apply it to compute neural networks.
Deep Learning, Scheduling, Parallel Programming, Inference, Performance Optimization, Heuristic Algorithm.
Hassan Osman Ali and Osama Rahmeh, Department of Information Security, Faculty of Computer Information Science (CIS), Higher Collages of Technology (HCT), Fujairah, UAE
As usually in every software project have to deal with change. Being able to effectively control and handle the proposed changes is crucial for allowing continued development of a software project to occur. To mitigate and control the change, developers must assess the risks related in doing the change. To understand the risk, the project manager must identify where the change will effect and the impact of the change entire the project. As usual, each change has its own risk, in smaller project will have fewer and manageable risk, but, larger project has higher level of risks. However, it can inundate yourself with too many changes request if you don’t take a focused approach that you can handle the risk. Some project managers are extremely slow moving, analytical types of project in which all requirements must be collected and assessed. On the other hand, some recent surveys reported that the project success rate has slightly better and increased over the last decade. This success is the outcome of describing a process and use of some available tools like requirements management tools. But, these tools are not mostly enough to handle and control the risks involve with the proposed change. Therefore, this paper focuses more on the factors that have high impact with the assessment and evaluation of the risks involve to the requested change.
Risks involve software changes, Software Change requirements, Risk management.
Shaher Ahmed, Mohamed Shekha, Suhaila Skran and Abdelrahman Bassyouny, Department of Mechatronics Engineering, Faculty of Engineering and Materials Science, The German University in Cairo, Egypt
The reduction of passenger journey time in an elevator system is an important goal in the lift industry. The major obstacle that prevents the optimization of the elevator dispatching is the uncertain traffic flow of passengers. In this paper, a comparison between the use of multiple Optimization Techniques such as Simulated Annealing (SA), Genetic Algorithm (GA), Particle Swarm Optimization Algorithm (PSO), and Whale Optimization Algorithm (WOA), are presented. A case study has been designed to analyze the functionality of the algorithms and to obtain a reasonable solution. To compare the results of the algorithms, performance indices are computed such as average and optimal fitness value in 5 runs in order to find the best algorithm for the elevator dispatching problem. The objective of this study is to reduce the average journey time for all passengers by computing a dispatching scheme; which is the output of the algorithms. The proposed technology would improve lift efficiency and provide a better user experience.
Stochastic Optimization, Elevator Dispatching Systems, Meta-Heuristics Optimization Techniques.
Lukas Jonathan Weber1, Alice Kirchheim2, Axel Zimmermann3, 1Department Mechanical Engineering, Helmut-Schmidt-University, Hamburg, Germany, 2Department Mechanical Engineering, Helmut-Schmidt-University, Hamburg, Germany, 3esz-partner Eber, Schwarzer, Zimmermann GbR, Kirchheim, Baden-Württemberg, Germany
The demand for accurate text mining tools to extract information of company based automotive warranty and goodwill (W&G) data is steadily increasing. The progress of the analytical competence of text mining methods for information extraction is among others based on the developments and insights of deep learning techniques applied in natural language processing (NLP). Directly applying NLP based architectures to automotive W&G text mining would wage to a significant performance loss due to different word distributions of general domain and W&G specific corpora. Therefore, labelled W&G training datasets are necessary to transform a general-domain language model in a specific-domain one to increase the performance in W&G text mining tasks. In this article, we describe a concept for adapting the generally pre-trained language model BERT  with the popular two-stage language model training approach in the automotive W&G context. For performance evaluation, we plan to use the common metrics recall, precision and F1-score.
Natural language processing, Domain-specific language models, BERT, Labelled domain-specific datasets, Automotive warranty and goodwill.