Jack Dongarra
Depei Qian

The 15th HPC Connection Workshop

Machine Intelligence and Supercomputing

April 27, 2017

National Supercomputing Center in Wuxi, Jiangsu, China

The HPC Connection Workshop is an international High Performance Computing summit organized by the Asia Supercomputer Community and Inspur Group. This event takes place three times a year: during the ASC in China, the ISC in Germany, and the SC in America. We are committed to the exploration of application innovation and current trends in emerging fields of science and engineering. During the workshop, top researchers and leading professionals from around the world gather together to discuss disruptive technologies and latest development in supercomputing.
Previous workshops http://asc-events.org/HPCCworkshopISC16.php   http://www.asc-events.org/HPC_Connection_Workshop_SC16.php






ASC Advisory Committee Chair
University of Tennessee
Oak Ridge National Laboratory

An Overview of High Performance Computing and the Design and Implementation of a Dense Linear Algebra Library
In this talk we will look at the current state of high performance computing and look at the next stage of extreme computing. With extreme computing, there will be fundamental changes in the character of linear algebra libraries. In this talk we will look at how extreme scale computing has caused algorithm and software developers to change their way of thinking on how to implement and program certain applications.


Professor, Beihang University, Sun Yat-sen University
Director of the Key Project on HPC, National High-Tech R&D Program


Professor Emeritus, Computer Science, University of Bonn

Machine Spaces for Universal Induction
In the sixties Solomonoff introduced a universal induction scheme which eventually would learn every computable pattern in a data sequence. Unfortunately, this generality comes at a price: the universal induction scheme is not computable, and within the asynchronous learning framework used by Solomonoff the trade-off between universality and effectivity is, in fact, unavoidable. But if one changes the asynchronous learning framework into a synchronous one, i.e., one within which the time scales of the learning system and the data generating process are coupled, this trade-off will vanish and effective universal induction becomes possible. Axioms and metrics for reference machine spaces are derived.


Professor, Waseda University
Founder of α-Expectation-Maximization Algorithm

Human-aware IoCT via Machine Learning and HPC
The IoT should not be merely a means of information labeling or communications. Processors in the IoT environment can cooperate towards the IoCT (Internet of Collaborative Things). Because each human is a node on the Internet, the machine side should provide human-aware provisions. The human side also learns to polish up its skill to use the IoCT environment. Thus, the cooperation of the human and the machine creates the next-generation IoCT. The computing performance becomes a key issue together with the communications. In this talk, the concept of the IoCT and human-aware examples of machine learning are exemplified.


Baidu Research


Deputy Director, National Supercomputing Center in Wuxi
Associate Professor, Tsinghua University

swDNN: A Library for Accelerating Deep Learning Applications on Sunway TaihuLight
The Sunway TaihuLight supercomputer is the world's first system with a peak performance greater than 100 PFlops, and a parallel scale of over 10 million cores. To explore the potential of training complex deep neural networks (DNNs) on other commercial chips rather than GPUs, we report our work on swDNN, which is a highly efficient library for accelerating deep learning applications on the newly announced world-leading supercomputer, Sunway TaihuLight. Targeting SW26010 processor, we derive a performance model that guides us in the process of identifying the most suitable approach for mapping the convolutional neural networks (CNNs) onto the 260 cores within the chip. By performing a systematic optimization that explores major factors, such as organization of convolution loops, blocking techniques, register data communication schemes, as well as reordering strategies for the two pipelines of instructions, we manage to achieve a double-precision performance over 1.6 Tflops for the convolution kernel, achieving 54% of the theoretical peak.


Panel Discussion

The workshop is free if you sign up before April 10. No other expenses will be covered due to high attendance. If there are any questions, please contact info@asc-events.org.



First Name *  
Last Name *  
Country *  
Email *  
Job Title *  
Organization *  
Occupation *
I would like to attend the HPC Connection Workshop on Apr 27
the ASC17 Awards Ceremony on Apr 28
* Required


Contact Us
Technical Support Yu Liu, Weiwei Wang techsupport@asc-events.org
Media Jie He media@asc-events.org
Collaboration Yan Yan executive.director@asc-events.org
General Information Dongzhi Kong info@asc-events.org


Copyright 2017 Asia Supercomputer Community. All Rights Reserved