SFB 876: Providing Information by Resource-Constrained Data Analysis

SFB 876研究中心汇集了数据分析（大数据分析、数据挖掘、机器学习、统计学）和网络物理系统（嵌入式系统）的研究领域，并增强了他们的方法，使分布式、动态的海量数据中的信息随时随地可用。资源限制源于需要以有效的方式处理大容量、大维度或高速数据。数据大小与计算资源之间的关系决定了算法所需的可扩展性。对于小型的、无处不在的设备来说，资源限制是显而易见的，但是对于大型计算设施来说，在非常大的、高维的、动态数据的情况下，它们也同样适用。虽然运行时是一个标准的考虑因素，但研究社区对内存、能源和通信等进一步资源的关注在最近的研究中已经集中起来。考虑到资源需求，研究了几种编程范例（数据流，MapReduce）和执行平台（FPGA， GPU）的数据分析算法。建立资源模型是为了正确评价资源，即能源消耗或通信。从这些角度来看，sfb876是一个先驱。SFB 876项目的协调如下。B类项目中的挑战来自本地的、移动的、无处不在的系统，如分布式嵌入式系统或小型病毒扫描器。C项目中的资源限制来自高维、动态和数据量，例如由天文学和粒子物理学产生的数据。A领域的总体项目包含了其他两个支柱。

The research center SFB 876 brings together the research fields of data analysis (Big Data Analytics, Data Mining, Machine Learning, Statistics) and Cyber-Physical Systems (Embedded Systems) and enhances their methods such that information from distributed, dynamic masses of data becomes available anytime and anywhere.Resource constraints arise from the need to handle large-volume, large-dimensional, or high-velocity data in an efficient way. The relation between the data size and the computing resources determines the necessary scalability of algorithms. For small, ubiquitous devices, the resource constraints are directly apparent, but they hold as well for large computing facilities in the case of extremely large, high-dimensional, dynamic data. Whereas runtime is a standard consideration, attention of the research community for further resources such as memory, energy, and communication has come to focus in recent research. Data analysis algorithms are investigated for several programming paradigms (data streams, MapReduce) and execution platforms (FPGA, GPU), taking their resource requirements into account. Resource models are established for the correct evaluation of resources, namely energy consumption or communication. In these perspectives, SFB 876 is a pioneer.Projects in SFB 876 are coordinated as follows. The challenges in the B projects stem from local, mobile, ubiquitous systems such as distributed embedded systems or small virus scanners. Resource constraints in the C projects arise from the high dimensionality, dynamics, and volume of the data, as are generated, for instance, by astro and particle physics. The overarching projects in the A domain encompass both other pillars.