How can we assure performance and dependability given the increasing complexity and data of an always-on connected world? Can we exploit power of tensor algebra to solve high-dimensional large-scale machine learning problems that such a world poses?

Cyber-physical systems (CPS) enable the physical world to merge with the virtual, leading to an internet of things, data, services, and people. This workshop will focus on the Internet of Things (IoT), i.e. devices, which have the capability to sense, communicate, and control. These devices are part of complex, dynamic, and distributed systems, such as in electricity or mobility networks. The various sensors enable them to capture multiple aspects of their surroundings in real-time. For example, phasor measurement units capture transient dynamics and evolving disturbances in the power system in high-resolution, in a synchronized manner, and in real-time. Another example is traffic networks, where a car today can deliver about 250 GB of data per hour from connected electronics such as weather sensors within the car, parking cameras and radars. Experts estimate that the IoT will consist of almost 50 billion objects by 2020. Big data computing frameworks, such as Hadoop, Spark or Storm currently form the basis for handling the massive amounts of data in batch and in stream. Advances in hardware such as many-core and heterogeneous architectures are also enabling factors. In order to enhance knowledge discovery, we believe that machine learning methods that can deal with multidimensional data are required to effectively extract information from this high-volume high-velocity sensor data. Crucial for the extraction of information is the format in which the data is represented.

The goal of the workshop is to explore tensor representations as the basis for machine learning solutions for the IoT. Tensors are algebraic objects which describe linear and multilinear relationships, and are represented as multidimensional arrays. They provide often a natural and compact representation for multidimensional data. In the recent years, tensor and machine learning communities, mainly active in the data-rich domains such as neuroscience, social network analysis, chemometrics, computer vision, knowledge graphs etc., have provided a solid research infrastructure reaching from the efficient routines for tensor calculus to methods of multi-way analysis to tensor decompositions for consistent and efficient estimation of parameters of the probabilistic models in search for convergence to globally optimal solutions. Big data necessitates optimized performance and space if latency and delay are to be minimized. These are also critical concerns in CPS, where real-time execution and security are desired. Great potential may exist for designing more efficient routines, which take into account the various computing architectures and resources that will co-exist in the heterogeneous IoT scenarios: complex processor/memory/network hierarchies and embedded processors.

This Dagstuhl Perspectives Workshop is planned as a catalyst and should attract experts from tensor and large scale data analysis, practitioners of hardware and software advancement for big data computing, as well as industrial stakeholders active in IoT/CPS. The resulting Tensor Computing for IoT manifesto will show the path to fill remaining gaps between the disciplines and outline a small number of concise action items that describe important directions of future research and investment. Topics of interest include:

• Limitations of the currently available models and algorithms for IoT data
• Employing tensor methods in conjunction with other methods for probabilistic modeling, deep learning
• Distributed data and computing models for multidimensional sensor data across heterogeneous architectures of multi-core cluster and embedded computing
• Universal algebra and calculus of indexing for optimized and verifiable composition of operations in an n-dimensional array/tensor algebra supporting the above models

• Evrim Acar (University of Copenhagen, DK) [dblp]
• Kareem Aggour (General Electric - Niskayuna, US) [dblp]
• Animashree Anandkumar (University of California - Irvine, US) [dblp]
• Rasmus Bro (University of Copenhagen, DK) [dblp]
• Ali Taylan Cemgil (Bogaziçi University - Istanbul, TR) [dblp]
• Edward Curry (National University of Ireland - Galway, IE) [dblp]
• Lieven De Lathauwer (KU Leuven, BE) [dblp]
• Hans Hagen (TU Kaiserslautern, DE) [dblp]
• Souleiman Hasan (National University of Ireland - Galway, IE) [dblp]
• Denis Krompaß (Siemens AG - München, DE) [dblp]
• Gerwald Lichtenberg (HAW - Hamburg, DE) [dblp]
• Benoit Meister (Reservoir Labs, Inc. - New York, US) [dblp]
• Morten Mørup (Technical University of Denmark - Lyngby, DK) [dblp]
• Lenore Mullin (University of Albany - SUNY, US) [dblp]
• Axel-Cyrille Ngonga Ngomo (Universität Leipzig, DE) [dblp]
• Ivan Oseledets (Skoltech - Scolkovo, RU) [dblp]
• Renato Pajarola (Universität Zürich, CH) [dblp]
• Evangelos Papalexakis (Carnegie Mellon University, US) [dblp]
• Christine Preisach (SAP SE - Walldorf, DE) [dblp]
• Achim Rettinger (KIT - Karlsruher Institut für Technologie, DE) [dblp]
• Sükran Sebnem Rusitschka (Siemens AG - München, DE) [dblp]
• Volker Tresp (Siemens AG - München, DE) [dblp]
• Bülent Yener (Rensselaer Polytechnic Institute - Troy, US) [dblp]