![](../images/_general/nice.jpg) |
The Fourth International Conference on Data Analytics
DATA ANALYTICS 2015
July 19 - 24, 2015 - Nice, France |
T1. 1:30pm -3:00pm
Random Number Generators with Multiple Streams for Parallel Computing: An Overview and Some Proposals
Prof. Dr. Pierre L'Ecuyer, Universite de Montreal, Canada
T2. 3:30pm - 5:00pm
Distributed Things: Technologies, Platforms, and Applications
Prof. Dr. Giorgio Delzanno, University of Genoa, Italy
T3. 5:30pm - 6:30pm
The First Rule of Marketing Analytics: Forget the Customer
Matt Hertig, Co-Founder, Alight Analytics, USA
DETAILS
T1. 1:30pm -3:00pm
Random Number Generators with Multiple Streams for Parallel Computing: An Overview and Some Proposals
Prof. Dr. Pierre L'Ecuyer, Universite de Montreal, Canada
Canada Research Chair in Stochastic Simulation and Optimization
DIRO, Universite de Montreal, Canada, and
Inria International Chair, Inria-Rennes, France
http://www.iro.umontreal.ca/lecuyer
In this tutorial talk, we examine the design of software libraries that can provide multiple streams of independent uniform random numbers for simulation in parallel computing environments. These multiple streams are typically dened as disjoint segments of the sequence of numbers produced by a single random number generator (RNG), which should behave approximately as the realizations of independent random variables uniformly distributed over the interval (0; 1) [2, 4, 5]. These numbers can be transformed appropriately to simulate random variables from other distributions, stochastic processes, and other types of random objects. Thousands or even millions of independent streams of random numbers are sometimes required in parallel computing applications. Multiple streams are also very convenient when running simulations on a single processor, for example to maintain proper synchronization when comparing similar systems with common random numbers (CRNs) and in simulation-based optimization via sample average approximation (SAA) [1, 3, 6, 7].
We consider different ways of providing streams and substreams. In the most traditional scheme, all the new streams must be created by a central manager that would reside at a single location (a host computer). Once created, the streams can be used anywhere. In an alternative scheme, new streams can be created from anywhere by transforming the state of an existing stream. Various implementations of each scheme will be discussed. We give special attention to parallel processing situations where each processor has a limited amount of fast-access private memory, such as for discrete graphical processing units (GPUs) and general-purpose GPUs (GPGPUs).
Finally, we introduce clRNG, an API and library for uniform random number generation in OpenCL. Streams of random numbers can be seen as virtual random number generators. They can be created on the host computer in unlimited numbers, and then used either on the host or on other computing devices by work items to generate random numbers. Each stream also has equally-spaced substreams, which are useful in certain settings. We provide examples showing the usefulness of streams and substreams in this context, and how the clRNG library can be used.
This talk is based on joint work with David Munger and Nabil Kemerchou.
References
[1] A. M. Law. Simulation Modeling and Analysis. McGraw-Hill, New York, fth edition, 2014.
[2] P. L'Ecuyer. Uniform random number generation. Annals of Operations Research, 53:77 /120, 1994.
[3] P. L'Ecuyer. Variance reduction's greatest hits. In Proceedings of the 2007 European Simulation and Modeling Conference, pages 5{12, Ghent, Belgium, 2007. EUROSIS.
[4] P. L'Ecuyer. Random number generation. In J. E. Gentle, W. Haerdle, and Y. Mori, editors, Handbook of Computational Statistics, pages 35{71. Springer-Verlag, Berlin, second
edition, 2012.
[5] P. L'Ecuyer, B. Oreshkin, and R. Simard. Random numbers for parallel computers: Requirements and methods, 2014. http://www.iro.umontreal.ca/~lecuyer/myftp/
papers/parallel-rng-imacs.pdf.
[6] P. L'Ecuyer, R. Simard, E. J. Chen, and W. D. Kelton. An object-oriented random-number package with many long streams and substreams. Operations Research, 50(6):1073{1075, 2002.
[7] A. Shapiro, D. Dentcheva, and A. Ruszczynski, editors. Lecture Notes on Stochastic Programming: Modeling and Theory. SIAM, Philadelphia, 2009.
T2. 3:30pm - 5:00pm
Distributed Things: Technologies, Platforms, and Applications
Prof. Dr. Giorgio Delzanno, University of Genoa, Italy
The Internet of Things (IoT) is a new paradigm for building distributed applications by connecting distributed objects using the Internet infrastructure as a Machine to Machine (M2M) communication media and using the cloud to maintain the resulting software architectures. IoT applications can be viewed as complex hardware and software systems composed by heterogeneous components like server stations, data centers, sensor devices, etc. Internet protocols are used to connect objects and common interface languages are used to make them interoperable. IoT platforms provide tools for building scalable applications combining data integration, analysis, synthesis, communication and visualization within a single framework. Furthermore, they provide cloud-based support tools for software maintenance, with high flexibility in the configuration of available services and data, in the update of existing software components and in the integration of new software applications and hardware devices.
IoT platforms are particularly useful for building applications for improving the quality of industrial, social and environmental services. In the tutorial, we will discuss technologies supporting the IoT philosophy, platforms designed to build scalable distributed systems, and possible applications of the resulting paradigm.
T3. 5:30pm - 6:30pm
The First Rule of Marketing Analytics: Forget the Customer
Matt Hertig, Co-Founder, Alight Analytics, USA
The customer decision journey is no longer linear – the method for measuring it shouldn’t be either. This session will challenge marketers to understand customer acquisition differently, how to construct a marketing measurement funnel, and how to demonstrate holistic results that drive performance and optimization.