Albert Bifet is a Senior Researcher at Huawei Noah's Ark Lab. He is the author of a book on Adaptive Stream Mining and Pattern Learning and Mining from Evolving Data Streams. He is one of the leaders of MOA and Apache SAMOA software environments for implementing algorithms and running experiments for online learning from evolving data streams. He is serving as Co-Chair of the Industrial track of ECML PKDD 2015, and as Co-Chair of BigMine (2015, 2014, 2013, 2012), and ACM SAC Data Streams Track (2016, 2015, 2014, 2013, 2012).
The lecture starts with illustrative examples that motivate big data mining, focusing in the continuous flow of data and time-evolving characteristics. We present the main characteristics in designing streaming algorithms, and illustrative examples of streaming algorithms for a set of data mining tasks, including clustering, classification, prediction, frequent pattern mining and novelty detection. The last part of the lecture, includes a software demonstration using MOA.
Christian Bockermann is a Ph.D. student at the TU Dortmund University. He received his diploma in Computer Science in 2007 and is involved in various projects with a strong focus on streaming data, ranging from IP-TV usage analytics over log-management to stream processing of telescope data. He is the creator of the streams framework, which focuses on the modeling of streaming applications from small devices (Raspberry PI, Android) to larger compute clusters.
Realtime processing of streaming data has become a key factor in modern Big Data based applications. This need has spawned numerous streaming engines that aim at scalable execution in large scale clusters, such as Apache Storm, Samza, Flink or MillWheel. On this quickly advancing topic, the big question is: What is the best choice of platform for the implementation of a streaming application?
The streams framework tackles this by modelling an streaming applications by their abstract data-flows. This component based approach aims at a platform independent implementation of custom code that allows for the execution of applications on different streaming engines -- ranging from mobile Android phones to Apache Storm clusters.
The talk gives an overview of the streams concepts and shows how to easily define applications based on existing components.
Andre C. P. L. F. Carvalho received his B.Sc. and M.Sc. degrees in Computer Science from the Universidade Federal de Pernambuco, Brazil. He received his Ph.D. degree in Electronic Engineering from the University of Kent, UK. Prof. Andre de Carvalho is Full Professor at the Department of Computer Science, Universidade de Sao Paulo, Brazil. He has published around 90 Journal and 200 Conference refereed papers. He has been involved in the organization of several conferences and journal special issues. His main interests are Machine Learning, Data Mining, Bioinformatics, Evolutionary Computation, Bioinspired Computing and Hybrid Intelligent Systems.
The lecture starts with illustrative examples that motivate big data mining, focusing in the continuous flow of data and time-evolving characteristics. We present the main characteristics in designing streaming algorithms, and illustrative examples of streaming algorithms for a set of data mining tasks, including clustering, classification, prediction, frequent pattern mining and novelty detection. The last part of the lecture, includes a software demonstration using MOA.
Michelangelo Ceci, Ph.D., is an assistant professor at the Dept of Computer Science, University of Bari, Italy. His main research interests are in data mining and machine learning. He was a visiting researcher at the University of Bristol (U.K.) and at the JSI (SLO). He has published more than 140 papers in refereed journals and conferences. He is the unit coordinator of the projects MAESTRA and Vi-POC. He has served in the Program Committee of many conferences, including: IEEE ICDM, IJCAI, ECMLPKDD (also as area chair), SIAM SDM, ISMIS, PAKDD and DS. He is member of the editorial boards of: IJSNM, IJDSN, IJDATS and JAIS. He has been the program co-chair of five workshops and DS2016, the organizing committee chair of SEBD 2007, member of the editorial committee of ``Intelligenza Artificiale'' and member of the editorial board of the ECMLPKDD 2014 and 2015 journal tracks.
This tutorial will introduce the different tasks of predicting structured outpputs, such as multi-target regression, multi-label classification and hierarchical multi-label classification. It will then present tree-based approaches to solving such tasks in the framework of predictive clustering. It will then cover recent extensions of these approaches towards learning from network data, semi-supervised learning and learning from data streams. It will conclude with a review of some practical applications of the presented methods in the areas of functional genomics, sensor networks, and text/multimedia annotation and retrieval.
Dr. Jian-Jia Chen joined Department of Informatics at TU Dortmund in Germany as Professor of Design Automation of Embedded Systems (Lehrstuhl 12) in April 2014. Before joining TU Dortmund, he was Junior professor in Department of Informatics in Karlsruhe Institute of Technology (KIT) from May 2010 to March 2014. He received his Ph.D. degree from Department of Computer Science and Information Engineering, National Taiwan University, Taiwan in 2006. He received his B.S. degree from the Department of Chemistry at National Taiwan University 2001. After finishing the compulsory civil service in Dec. 2007, between Jan. 2008 and April 2010, he was a postdoc researcher at Computer Engineering and Networks Laboratory (TIK) in Swiss Federal Institute of Technology (ETH) Zurich, Switzerland. His research interests include real-time systems, embedded systems, cyber-physical systems, energy-efficient and power-aware designs. He received Best Paper Awards from ACM Symposium on Applied Computing (SAC) in 2009 and IEEE International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA) in 2005, 2013, and SmartGreen in 2014.
To deal with a large variety of workloads in different application domains in cyber physical systems, a number of expressive task models have been developed to jointly consider the flexibility and dynamics of communication, control, and computation capability. This results in a large design space for verifications and optimizations. This talk will first review these models and explain the potential solutions for such complicated execution models that involve certain dynamics with the environment. We will go through several examples by considering the environmental-induced mode changes, the accelerations and communications with other components, and the potential of energy reductions. We will then briefly go through a general schedulability analysis framework to verify the timeliness of such complicated execution models. The talk will end by exploring potential energy-efficient and power-aware designs for multi-core systems with different power and energy characteristics among the cores.
Sašo Džeroski is a scientific councillor at the Jozef Stefan Institute and the Centre of Excellence for Integrated Approaches in Chemistry and Biology of Proteins, both in Ljubljana, Slovenia. He is also a full professor at the Jozef Stefan International Postgraduate School. His research is mainly in the area of machine learning and data mining (including structured output prediction and automated modeling of dynamic systems) and their applications (mainly in environmental sciences, incl. ecology, and life sciences, incl. systems biology).
He has organized many scientific events, most recently two workshops on Machine Learning in Systems Biology and the International Conference on Discovery Science. He is co-author/co-editor of more than ten books/volumes, including "Inductive Logic Programming", "Relational Data Mining", "Learning Language in Logic", "Computational Discovery of Scientific Knowledge" and "Inductive Databases and Constraint-Based Data Mining". He has participated in many international research projects (mostly EU-funded) and coordinated two of them in the past: He is currently the coordinator of the FET XTrack project MAESTRA (Learning from Massive, Incompletely annotated, and Structured Data) and one of the principal investigators in the FET Flagship Human Brain Project.
This tutorial will introduce the different tasks of predicting structured outpputs, such as multi-target regression, multi-label classification and hierarchical multi-label classification. It will then present tree-based approaches to solving such tasks in the framework of predictive clustering. It will then cover recent extensions of these approaches towards learning from network data, semi-supervised learning and learning from data streams. It will conclude with a review of some practical applications of the presented methods in the areas of functional genomics, sensor networks, and text/multimedia annotation and retrieval.
João Gama received his Licenciado degree from the Fac. of Engineering of the University of Porto, Portugal. In 2000 he received his Ph.D. degree in Computer Science from the Faculty of Sciences of the same University. He joined the Faculty of Economy where he holds the position of Associate Professor, He is also a senior researcher at LIAAD, a group belonging to INESC Porto. He has worked in projects and authored papers in areas related to machine learning, data streams and adaptive learning systems and is a member of the editorial board of international journals in his area of expertise.
The lecture starts with illustrative examples that motivate big data mining, focusing in the continuous flow of data and time-evolving characteristics. We present the main characteristics in designing streaming algorithms, and illustrative examples of streaming algorithms for a set of data mining tasks, including clustering, classification, prediction, frequent pattern mining and novelty detection. The last part of the lecture, includes a software demonstration using MOA.
Ricard Gavaldà is a professor at the Department of Computer Science of Universitat Politècnica de Catalunya (UPC), in Barcelona. With a background on complexity and algorithms, his research during the last decade has focused on algorithmic aspects of machine learning and data mining, with emphasis on scalability. In particular, he is currently working on data stream mining and on methods for inferring latent variable models. He is also interested on the practical application of these techniques to domains such as autonomic computing, healthcare, social network analysis, and urban mobility. He has published over 80 conference or journal papers, supervised 7 Ph.D. theses, and contributed to open-source software platforms such as MOA (Massive Online Analysis).
This talk will cover some of the algorithmic techniques for efficient computing on data streams. They are usually embodied in "sketches" that use low time per stream item, sublinear memory, and can provide anytime answers. We will concentrate on those that are most likely to be useful in the design of higher-level data mining and machine learning algorithms on streams. Many of these algorithms are randomized and approximate, so the talk will also present the main probabilistic tools for designing and analyzing them.
Fosca Giannotti is a director of research at the Information Science and Technology Institute of the National Research Council, ISTI-CNR, Pisa, Italy. Her current research interests include spatio-temporal data mining, privacy preserving data mining, social network analysis, data mining query languages. She been the coordinator of various European research projects, including the FP6-IST project GeoPKDD and currently is coordinator of the H2020 RI Sobigdata.eu. She is the author of more than one hundred publications and served in the scientific committee of the main conferences in the area of Databases and Data Mining. She chaired ECML/PKDD 2004, the European Conf. on Machine Learning and Knowledge Discovery in Data Bases, and ICDM 2008, the IEEE Int. Conf. on Data Mining. Fosca Giannotti co-leads the KDD Lab – Knowledge Discovery and Data Mining Laboratory2 – a joint research initiative of the University of Pisa and ISTI- CNR.
The wireless networks that surround us, as a by-product of their normal operations, allow for sensing and collecting massive repositories of spatio-temporal data, such as the call detail records from mobile phones and the GPS tracks from car navigation devices, which represent society-wide proxies of human mobile activities. These big mobility data provide an unprecedented powerful social microscope, which helps us understand human mobility, and discover the hidden patterns and profiles that characterize the trajectories we follow during our daily activity. The lecture will illustrate the basic methods of mobility data mining, designed to extract from the big mobility data the patterns of collective movement behavior, i.e., discover the subgroups of travelers characterized by a common purpose, and the profiles of individual movement activity, i.e., characterize the routine mobility of each traveller. We also present how mobility data mining can be combined with complex network analysis to address fascinating new questions, such as how to discover the geographical borders that emerge from the network of flows between any two zones in a territory, and how to measure to what extent the mobility patterns shape and impact the social networks we inhabit.
Aristides Gionis is the director of the Algorithmic Data Analysis (ADA) programme in HIIT and he leads the Data Mining group in Aalto University. Previously he has been a senior research scientist in Yahoo! Research. He received his PhD from the Computer Science department of Stanford University in 2003. He is currently serving as an associate editor in the IEEE Transactions of Knowledge and Data Engineering (TKDE) and the ACM Transactions of Knowledge Discovery from Data (TKDD), and as a managing editor in Internet Mathematics. He has served in the PC of numerous premium conferences, including being the PC chair for WSDM 2013 and ECML PKDD 2010.
A lot of research in graph mining and social-network analysis has been devoted to the problem of community discovery. Finding dense subgraphs is an important algorithmic tool for discovering meaningful communities in graphs. In this tutorial we will present key concepts in dense-subgraph discovery. We will review different problem formulations, we will discuss their complexity, and present efficient algorithms. We will also discuss a number of variants to the main theme: finding dense subgraphs with respect to reference nodes, taking into account node weights, or temporal information.
Carlos Gómez-Uribe is the Vice President of Product Innovation at Netflix, leading the teams responsible for the video recommendation and search algorithms that connect Netflix members with movies and TV shows to enjoy. Carlos is a scientist and engineer turned product leader with years of experience applying mathematical and statistical modeling techniques in a variety of industries. Prior to Netflix, Carlos worked at Google, McKinsey & Company, Merrill Lynch and Analog Devices. His education includes a PhD in Medical and Electrical Engineering, a Master’s Degree in Electrical Engineering and Computer Science, and Bachelor degrees in Mathematics and in Electrical Engineering and Computer Science, all from the Massachusetts Institute of Technology.
This lecture will revolve around the innovation process we use at Netflix to improve our product, particularly the recommendation and search algorithms that help our members find movies and TV shows to watch. Our recommendation and search algorithms are key to retaining our members, so improving them has been a top priority for over a decade. Our innovation process, at its core, is an application of the scientific method, relying heavily on experimentation, very large data sets, carefully chosen metrics, and advanced statistical and mathematical models. We will discuss how our innovation process, despite its challenges and limitations, has lead to a remarkable evolution in our recommendation and search algorithms. We will also give an overview of the various algorithms that make up our recommender system, and will dive into some general mathematical techniques to produce recommendations, such as matrix factorization approaches.
Estevam R. Hruschka Jr. has received his Ph.D. degree in Computational Systems from Federal University of Rio de Janeiro, Brazil, 2003. He has been ”young research fellow” at FAPESP (Sao Paulo state research agency, Brazil) and, currently, he is ”research fellow” at CNPq (Brazilian research agency). Also, he is associate professor at Federal University of Sao Carlos (UFSCar), Brazil. Professor Estevam was visiting professor (2008-2010) at Carnegie Mellon University and has been working with Professor Tom Mitchell and Professor William Cohen (at the Carnegie Mellon Read the Web project group - http://rtw.ml.cmu.edu/rtw/people) since the beginning of the ReadTheWeb project (in 2008), helping to build the Never- Ending Language Learning (NELL) system that has started running on January 2010. His main research interests are never-ending learning, machine learning, probabilistic graphical models and natural language understanding. He has worked with many international research teams collaborating with research groups from universities and also from companies.
The Web is inundated with information in many different formats including semi-structured and unstructured data. Machine Reading is a research area that aims to build systems that can read natural-language-based information, ex- tracting knowledge and storing it into knowledge bases. Thus, Machine Reading systems are developed to produce language- understanding technology that will automatically process text in affordable time. In this tutorial the idea of automatically reading the Web using Machine Reading techniques will be explored, and discussions on how far have we came will be conducted.
Four of the most successful Machine Reading approaches in- tended to Read the Web (namely KnowItAll, Yago, NELL and DBPedia systems) will be presented and discussed. The principles, the subtleties, as well as current results of each approach will be addressed. On-line resources (from each approach) will be explored and the future directions in each system will be pointed out. YAGO, KnowItAll, NELL and DBPedia are not the only research efforts focusing on Reading the Web. They were selected, to be presented in this tutorial, because they show four different and very relevant approaches to this problem, but it does not mean they are the only relevant approaches at all. In spite of mainly focusing on the four aforementioned systems, some other independent contributions on the Read the Web idea will be mentioned and pointed out as related works.
Dr. Peter Marwedel studied physics at the University of Kiel, Germany. He received his PhD in that subject in 1974. As a post-doc, he published some of the first papers on high-level synthesis and retargetable compilation in the context of the MIMOLA hardware description language. In 1987, his habilitation thesis in computer science (a thesis required for becoming a professor) was accepted. Since 1989, he is holding a chair for computer engineering and embedded systems at the computer science department of TU Dortmund. He is also chairing ICD, a local spin-off of TU Dortmund. His research interests include design automation for embedded systems, in particular the generation of efficient embedded software. Focus is on energy efficiency and timing predictability. Since 2001, Dr. Marwedel published papers on energy-efficient software and compiler-supported use of scratchpad memories. He is the author of one of the few textbooks on embedded systems. Since 2011, he is the vice-chair of the collaborative research center SFB 876, aiming at resource-efficient analysis of large data sets. Dr. Marwedel is an IEEE Fellow.
The term cyber-physical systems characterizes the integration of information and communication technologies (ICT) with their physical environment. This integration results in a huge potential for the development of intelligent systems in a large set of industrial sectors. The potential will be covered in the first part of the talk. Sectors comprise industrial automation (industry 4.0), traffic, consumer devices, the smart grid, the health sector, urban living and computer-based analysis in science and engineering. A multitude of goals can be supported in this way, e.g., the availability of higher standards of living, higher efficiency of many processes, the generation of knowledge and safety for the society. However, the realization of this integration implies manifold challenges.
Challenges covered in the second part of this talk include security, timing, safety, reliability, energy efficiency, interfacing, and the discovery of information in huge amounts of data. Also, the inherent multidisciplinarity poses challenges for knowledge acquisition and application. In the third and final part of the talk, we will present some of our contributions addressing these issues. These contributions techniques for improving the energy efficiency include an integration of a timing model into the code generation process, tradeoffs between timeliness and reliability, and approaches for education crossing boundaries of involved disciplines.
Dr. Peter Marwedel studied physics at the University of Kiel, Germany. He received his PhD in that subject in 1974. As a post-doc, he published some of the first papers on high-level synthesis and retargetable compilation in the context of the MIMOLA hardware description language. In 1987, his habilitation thesis in computer science (a thesis required for becoming a professor) was accepted. Since 1989, he is holding a chair for computer engineering and embedded systems at the computer science department of TU Dortmund. He is also chairing ICD, a local spin-off of TU Dortmund. His research interests include design automation for embedded systems, in particular the generation of efficient embedded software. Focus is on energy efficiency and timing predictability. Since 2001, Dr. Marwedel published papers on energy-efficient software and compiler-supported use of scratchpad memories. He is the author of one of the few textbooks on embedded systems. Since 2011, he is the vice-chair of the collaborative research center SFB 876, aiming at resource-efficient analysis of large data sets. Dr. Marwedel is an IEEE Fellow.
To deal with a large variety of workloads in different application domains in cyber physical systems, a number of expressive task models have been developed to jointly consider the flexibility and dynamics of communication, control, and computation capability. This results in a large design space for verifications and optimizations. This talk will first review these models and explain the potential solutions for such complicated execution models that involve certain dynamics with the environment. We will go through several examples by considering the environmental-induced mode changes, the accelerations and communications with other components, and the potential of energy reductions. We will then briefly go through a general schedulability analysis framework to verify the timeliness of such complicated execution models. The talk will end by exploring potential energy-efficient and power-aware designs for multi-core systems with different power and energy characteristics among the cores.
Nico Piatkowski is a PhD candidate at the artificial intelligence group of the TU Dortmund University, Germany. His main research is about resource efficient machine learning algorithms with a strong focus on probabilistic graphical models, regularization, approximation theory, parallel algorithms and complexity in statistical modeling. Nico studied at TU Dortmund where he received a diploma in computer science in 2011 and a B.Sc. in economics in 2013.
Uncertainty is inherent in almost all real-world applications since many important aspects of a system can only be partially observed or can not be observed at all. Probability theory allows us to model this imperfection by collections of interdependent random variables, reflecting estimated beliefs about the different possible states of the world. While probability theory has existed for more than three centuries, our ability to use it effectively on big problems is quite recent, and is due largely to the development of a framework known as Probabilistic Graphical Models (PGMs).
In this lecture, I will illustrate the basics of the PGM representation (like directed and undirected models, plate notation and factorization) and how the corresponding graphical structure and model parameters can be estimated. Multivariate Gaussian and multinomial distributions are related to this framework, complexity theoretical issues and approximation algorithms are discussed and some inherent limitations of common approaches are explained.
Jesse Read studied computer science and obtained his Ph.D. in 2010 in the machine learning group at the University of Waikato in New Zealand. He then began post-doctoral research in the University Carlos III of Madrid. Since 2013 he has been at Aalto University in Helsinki. Aside from multi-label classification, his research interests include learning from sequential data, evolving data streams, and working with applications of sensory data.
In multi-label classification, each data instance may be associated with multiple class labels, as opposed to a single class. This context arises naturally in many (if not most) domains. Text categorisation and image labelling are just two examples. Work in multi-label learning has proliferated in recent years in the academic literature. The main challenge is to model and deal with label dependence, and to do so efficiently enough for scaling up to the domain of interest. This tutorial reviews some important developments in multi-label classification, and their connection to related areas such as structured-output prediction. These developments include methods based on data transformation, probabilistic graphical models, and neural networks. A number of example applications will also be reviewed.
Salvatore Rinzivillo is researcher at the Information Science and Technology Institute of the National Research Council, ISTI-CNR, Pisa, Italy. He had his PhD from Computer Science Department of University of Pisa in 2006. His current research interests include smobility data mining data mining, social network analysis, and visual analytics. He has been involved in several European projects in the FET program and he had the responsibility of the workpackage "Big data driven theories of mobility demand" of the FP7 project DataSIM: DATA science for SIMulating the era of electric vehicles. One of the major achievements has been the contribute to understanding and modeling human mobility with the paper "Returners and Explorers: Dichotomy in human mobility" accepted for publication by Nature.
The wireless networks that surround us, as a by-product of their normal operations, allow for sensing and collecting massive repositories of spatio-temporal data, such as the call detail records from mobile phones and the GPS tracks from car navigation devices, which represent society-wide proxies of human mobile activities. These big mobility data provide an unprecedented powerful social microscope, which helps us understand human mobility, and discover the hidden patterns and profiles that characterize the trajectories we follow during our daily activity. The lecture will illustrate the basic methods of mobility data mining, designed to extract from the big mobility data the patterns of collective movement behavior, i.e., discover the subgroups of travelers characterized by a common purpose, and the profiles of individual movement activity, i.e., characterize the routine mobility of each traveller. We also present how mobility data mining can be combined with complex network analysis to address fascinating new questions, such as how to discover the geographical borders that emerge from the network of flows between any two zones in a territory, and how to measure to what extent the mobility patterns shape and impact the social networks we inhabit.
Sebastian Schelter is a PhD student at the Database Systems and Information Management Group (DIMA) of TU Berlin with Prof. Volker Markl. His research aims at improving the technology for performing large scale data analysis on parallel processing platforms. Use case-wise, the focus is on enabling Collaborative Filtering with billions of interactions and Graph Mining on graphs with billions of vertices and edges. Furthermore, Sebastian is part of the developing team of Apache Flink (formerly Stratosphere), a database-inspired parallel processing stack for large scale analytics. The system is part of a research project lead by the group at TU Berlin. Sebastian is also engaged in Open Source as a member of the Apache Software Foundation, where he is a committer and PMC member in the Mahout, Giraph and Flink projects.
An overview of his research work can be found at http://ssc.io/research
The talk will provide an overview over Apache Flink, a next-generation large scale data analysis plaform. Flink unifies technology from the world of parallel databases with MapReduce-like systems. An overview over Flink can be found at https://flink.apache.org
Christian Sohler studied Computer Science at Saarland University and obtained his Ph.D. from the University of Paderborn. Since 2009 he is associate professor for complexity theory and efficient algorithms at the Technical University of Dortmund. His main research areas are randomized algorithms, sublinear time algorithms, streaming algorithms, clustering, and computational geometry.
Can we predict from the structure of social networks of a country whether it is a democrathy or a totalitarian state? Is it possible to recognize schizophrenia by analyzing the network structure of the neurons of the human brain? If we would like to answer questions like the ones above we need to analyze the structure of hundreds of very large networks. Since graph isomorphism is not known to be efficiently solvable, we can only describe the structure of networks ''approximately'', for example, by transforming them into feature vectors. If we want to follow this approach for the above problems, we need to obtain a decent number of features for hundreds of very large networks. Even if we can compute each feature in linear time, this is hardly doable. Thus we need a different approach: Approximating the structure of graphs by random sampling.
In my talk I will explain how random walks can be used to provably approximate some graph properties (like expansion, bipartiteness and clusterability). I will use the framework of property testing that allows to study random sampling processes in graphs in a theoretical setting.
Marco Stolpe's current main interest is in the development of communication-efficient distributed in-network algorithms for strongly resource-constraint systems like wireless sensor networks. Furthermore, he has worked on the problem of learning from label proportions and the classification of time-related sensor data. He has a six year expertise in the creation of data analysis processes with RapidMiner. As a Ph.D. student of Katharina Morik at TU Dortmund, he is mainly involved with Collaborative Research Center SFB 876, project B3, that is concerned with data mining on sensor data of automated processes.
RapidMiner is regularly voted the most used data mining tool in the yearly KDNuggets.com survey. The software consists of a large tool box of operators which support the whole data analysis chain, from reading and preprocessing data, over modeling, feature selection and parameter optimization, to evaluation and reporting. A leading paradigm is to provide industrial-strength functionality to end users with moderate programming knowledge in a code-free graphical user interface.
Data analysis in research and academia, however, can deviate much from the standard processes created by RapidMiner's powerful wizards or being available in its samples directory. Especially the preprocessing of large datasets often requires innovative solutions beyond RapidMiner's well-known set of operators. This two hour tutorial therefore presents RapidMiner's standard functionality only shortly, but then dives deeply into lesser known areas like the creation of new data tables and columns, nominal mappings, macros, in-memory objects, loops, branches, the set data operator and the execution and parallelization of subprocesses. The interaction between such operators is exemplarily demonstrated and examined in a real-world case study of preprocessing sensor data from interlinked production processes.
Jens Teubner is leading the Databases and Information Systems Group at TU Dortmund in Germany. His main research interest is data processing on modern hardware platforms, including FPGAs, multi-core processors, and hardware-accelerated networks. Previously, Jens Teubner was a postdoctoral researcher at ETH Zurich (20082013) and IBM Research (20072008). He holds a PhD in Computer Science from TU Müchen (Munich, Germany) and an M.S. degree in Physics from the University of Konstanz in Germany.
The famous "Moore's Law" just turned 50. But while the exponential growth of hardware complexity still prevails, turning it into faster application software is becoming harder and harder. To unleash the potential of modern hardware, algorithms must be carefully tuned to make profitable use of deep cache hierarchies, thread- and data-level parallelism, novel instruction sets, and modern architectures like NUMA configurations.
In the lecture, I will illustrate what the hardware changes mean to data-intensive applications. We will start by looking into the effects of caches, and we will see how data layouts and access patterns can be tuned to maximize caching efficiency. Parallelism is another important ingredient to leverage the potential of modern hardware. But it comes with inconvenient side effects, such as locality and NUMA effects. By example of database algorithms, I will show how algorithms can be designed in a NUMA-efficient and scalable way. Finally, I will give a glimpse into technology beyond mainstream multi-core processors and show how and why co-processors like graphics processors or FPGAs are different.
Connect