Two-stage optimization for machine learning workflow

Machine learning techniques play a preponderant role in dealing with massive amount of data and are employed in almost every possible domain. Building a high quality machine learning model to be deployed in production is a challenging task, from both, the subject matter experts and the machine learning practitioners.For a broader adoption and scalability of machine learning systems, the construction and configuration of machine learning workflow need to gain in automation. In the last few years, several techniques have been developed in this direction, known as AutoML.In this paper, we present a two-stage optimization process to build data pipelines and configure machine learning algorithms. First, we study the impact of data pipelines compared to algorithm configuration in order to show the importance of data preprocessing over hyperparameter tuning. The second part presents policies to efficiently allocate search time between data pipeline construction and algorithm configuration. Those policies are agnostic from the metaoptimizer. Last, we present a metric to determine if a data pipeline is specific or independent from the algorithm, enabling fine-grain pipeline pruning and meta-learning for the coldstart problem.     

A benchmark data set for two-phase Coriolis metering

For more than a decade there has been growing interest in the use of Coriolis mass flow metering applied to two-phase (gas/liquid) and multiphase (oil/water/gas) conditions. It is well-established that the mass flow and density measurements generated from multiphase flows are subject to large errors, and a variety of physical models and correction techniques have been proposed to explain and/or to compensate for these errors. One difficulty is the absence of a common basis for comparing correction techniques, because different flowtube designs and configurations, as well as liquid and gas properties, may result in quite different error curves. Furthermore, some researchers with interests in the modelling aspects of the field may not have suitable multiphase laboratory facilities to generate their own data sets. This paper offers a small data set that may be used by researchers as a benchmark i.e. a common data set for comparing correction techniques. The data set was collected at the UK National Flow Laboratory TUV-NEL, using air and a viscous oil, and provides experimental points over a wide flow range (8:1 turndown) and with Gas Volume Fraction (GVF) values up to 60%. As a first investigation using the benchmark data set, we consider how data sparsity (i.e. the flow rate and GVF spacing in the experimental grid) affects the accuracy of a correction model. A range of neural network models are evaluated, based on different subsets of the benchmark data set. The data set and some exemplary code are provided with the paper. Additional data sets are available on a web site created to support this initiative.     

面向数据流结构的指令映射优化方法

在高性能计算领域,数据流是一类重要的计算结构,也在很多实际场景表现出很好的性能和适用性。在数据流计算模式中,程序是以数据流图来表示的,数据流计算中一个关键的问题是如何将数据流图映射到多个执行单元上。通过分析现有数据流结构的指令映射方法及其不足,提出了基于数据流结构的新型指令映射优化方法。主要是根据多地址共享数据包的特性对指令映射方法进行优化,延迟多地址共享数据路由包的拆分,减少网络拥堵。     

基于强化学习的数据驱动最优镇定控制及仿真

利用Q-学习算法,针对模型未知只有数据可用的非线性被控对象,解决最优镇定控制问题.由于状态空间和控制空间的连续性,Q-学习只能以近似的方式实现.因此,文中提出的近似Q-学习算法只能获得一个次优控制器.尽管求得的控制器只是次优,但是仿真研究表明,对于强非线性被控对象,相比线性二次型调节器和深度确定性梯度下降方法,文中方法的闭环吸引域更宽广,实际指标函数也更小.     

A design method of data analytics process for condition based maintenance

Many condition monitoring systems use data analytics processes such as anomaly detection to understand machine conditions. Such data analytics processes have been designed by data scientists. However, domain knowledge is indispensable for designing the process, and data scientists have difficulty in acquiring such knowledge from domain engineers. This paper proposes a design method of data analytics processes and an engineering tool. In this method, data scientists propose hypotheses about each step of the process to domain engineers. Then, data scientists update the process on the basis of feedback from domain engineers. The engineering tool helps data scientists to interact with domain engineers.     

Ontology-based approach for the provision of simulation knowledge acquired by Data and Text Mining processes

Numerical simulation techniques such as Finite Element Analyses are essential in today's engineering design practices. However, comprehensive knowledge is required for the setup of reliable simulations to verify strength and further product properties. Due to limited capacities, design-accompanying simulations are performed too rarely by experienced simulation engineers. Therefore, product models are not sufficiently verified or the simulations lead to wrong design decisions, if they are applied by less experienced users. This results in belated redesigns of already detailed product models and to highly cost- and time-intensive iterations in product development.Thus, in order to support less experienced simulation users in setting up reliable Finite Element Analyses, a novel ontology-based approach is presented. The knowledge management tools developed on the basis of this approach allow an automated acquisition and target-oriented provision of necessary simulation knowledge. This knowledge is acquired from existing simulation models and text-based documentations from previous product developments by Text and Data Mining. By offering support to less experienced simulation users, the presented approach may finally lead to a more efficient and extensive application of reliable FEA in product development.     

JRF33铁路车号识别装置在矿车计数系统的应用

针对采厂矿车翻卸人工计数可靠性低、安全风险大等问题，提出一种矿车翻卸自动计数系统设计方案，以JRF33铁路车号识别装置为载体，以RS485为通讯协议，对采集数据进行统计分析。结果显示，计数系统运行良好，翻卸数据采集及时、准确、完整。系统对采矿原料运输数据统计、分析应用和生产指导有一定的指导意义。     

Digital twin-based assembly data management and process traceability for complex products

Complex products such as satellites, missiles, and aircraft typically have demanding requirements for dynamic data management and process traceability. The assembly process for these complex products involves high complexity, strong dynamics, many uncertainties, and frequent rework and repair, especially in the model development stage. Achieving assembly data management and process traceability for complex products has always been a challenge. A recently proposed solution involves one-to-one mapping of the corresponding physical entity, also known as the digital twin method. This paper proposes a digital twin-based assembly data management and process traceability approach for complex products. First, the dynamic evolutionary process of complex product assembly data was analyzed from three dimensions: granularity, period and version. Then, a framework of digital twin-based assembly data management and process traceability for complex products was constructed. Some core techniques are: 1) workflow-based product assembly data organization and version management; 2) synchronous modeling of the product assembly process based on digital twin; and 3) hierarchical management and traceability of product assembly data based on digital twin. On this basis, an algorithm flowchart for generating a product assembly data package was created, which includes product assembly data management, assembly process traceability, and generation of a product assembly data package. Furthermore, the Digital Twin-based Assembly Process Management and Control System (DT-APMCS) was designed to verify the efficiency of the proposed approach. Some aerospace-related assembly enterprises are currently using DT-APMCS and achieving satisfactory results. Finally, a summary of our work is given, and the future research work is also discussed.     

Prediction methods and applications in the science of science: A survey

Science of science has become a popular topic that attracts great attentions from the research community. The development of data analytics technologies and the readily available scholarly data enable the exploration of data-driven prediction, which plays a pivotal role in finding the trend of scientific impact. In this paper, we analyse methods and applications in data-driven prediction in the science of science, and discuss their significance. First, we introduce the background and review the current state of the science of science. Second, we review data-driven prediction based on paper citation count, and investigate research issues in this area. Then, we discuss methods to predict scholar impact, and we analyse different approaches to promote the scholarly collaboration in the collaboration network. This paper also discusses open issues and existing challenges, and suggests potential research directions.     

Profiling of knowledge organisation systems for the annotation of Linked Data cultural resources

The importance of cultural heritage for supporting the knowledge economy has promoted its digitisation and online publication. Many cultural heritage repositories have published millions of digitised items using semantic web technologies and Linked Data approaches. These repositories frequently use knowledge organisation systems to classify the resources, but the domain heterogeneity makes it difficult to know if they are the most suitable ones. This paper describes the process used to discover and profile the knowledge organisation systems used in the cultural heritage domain. Additionally, for those knowledge organisation systems with a thesaurus-like structure, a detailed quality analysis is performed. The results of this analysis can be used as a key factor for the selection of knowledge organisation systems in classification tasks.