Tutorials Programme
Tutorials Programme
Schedule
Tutorial 1:
OpenPOWER and Power9Accelerate your AI training with IBM | AI + HPC Cloud Super Computer in the planet
- Date: Tue, March 24
Time: 1.30 pm to 4.30 pm
Limited access: 50 participants – first come first served - Brief agenda:
-
- Power 9/OpenPOWER Features
- WML features
- Xilinx U50 and Edge Compute
- Advanced accelerator features example OpenCAPI and Use cases
- Q and A, Way forward
- IBM has been working with leading AI partners Google, nVidia and Xilinx to bring the latest AI technology to many Universities worldwide, allowing student researchers and faculties to use AI supercomputers to learn and optimize AI applications in distributed environment. This has helped establish students’ overall understanding of artificial intelligence from the perspective of infrastructure and computing power, and cultivate talents with HPC+AI capabilities.
- Learn about Artificial Intelligence and gather the latest insights from pioneers in the industry, leveraging the POWER9 systems also used in the world’s largest supercomputer
- Learn about POWER9/OpenPOWER systems
- Discover advances in deep learning tools and techniques
- Learn how to use OpenPOWER systems and PowerAI tools to do your AI projects
- Go deeper with those who have especially challenging projects
- Presenters:
- Florin Manila
Bruno Mesnet, CAPI SNAP Proof of Concepts Team leader
Jens Stapelfeldt - About presenters:
- Florin leads enterprise transformation by adopting of High-Performance Computing, Distributed Deep Learning and Emerging Technologies in order to help them to rapidly transform the way businesses operate, solve problems, and gain competitive advantage. He is responsible for performance, availability, and scalability of Cognitive Systems infrastructure. It has created IBM Distributed Deep Learning reference
architecture as well as important industry blueprints of applied AI including edge fabric. His client experience covers EU commercial and
government civil and defense agencies. Is passionate about in-memory computing and Spiking Neural Networks.Bruno is an IBM Systems Technical lead engineer. He focuses on enablement and Proof of Concept of all applications that can be
accelerated. He has been promoting around Europe and US the SNAP concept. Based in Montpellier, France, Bruno enjoys easing user’s experience: “the simpler, the better”.Jens works as business development manager at xilinx for the global data centre group in europe. He has a 30 years background in electronics and the semiconductor space. Jens worked as arm trainer, senior feld application engineer and technical sales lead for doulos, texas instruments and in the last 5 years for xilinx. He has a masters in microelectronic and computer science and 2018 finished his part time mba and look onto the european ai start-up landscape. Furthermore jens is teaching on a local university (phwt) as part time professor for computer science and electronic design.
How to join?
During general registration for the conference. The tutorial will be conducted onlline – participants will receive a link to WebEx platform.
The following tutorials will be conducted in Warsaw in future times
Tutorial 2:
Advanced scientific visualization with VisNow platform
- Start time: 14:00
- Brief agenda:
-
- Introduction to Scientific Visualization and Visual Analysis.
- Visualization systems and paradigms
- Generic data structures
- Introduction to VisNow
- Hands-on Session #1 – 2D data visualization
- Hands-on Session #2 – 3D data visualization
- Hands-on Session #3 – Vector data visualization
- Hands-on Session #4 – Unstructured data visualization.
- Presenters:
- Jędrzej Nowosielski, PhD, Interdisciplinary Centre for Mathematical and Computational Modelling University of Warsaw, Poland
Krzysztof Nowiński, PhD, Interdisciplinary Centre for Mathematical and Computational Modelling University of Warsaw,
Poland - Abstract:
- Visual analysis is one of the most powerful tools for data exploration and interpretation. It takes advantage of visualization techniques and allows scientists to work with their research data in interactive and intuitive way. In today’s HPC environment and Big Data era, data analysis techniques, together with visualization, gain on importance. However, the amounts of data and the sizes of single datasets impose the need for adequate software tools. In this tutorial we will address this problem by providing participants with strong tool for data processing, visualization and visual analysis – VisNow, an open source generic platform based on data flow paradigm. The goal of this tutorial is to introduce the audience to the concept of visual analysis, show basic ideas of scientific visualization and to go step-by-step through several case studies in hands-on sessions based on our platform. Problems of visualization of common HPC data structures, including 2-D and 3-D, scalar and vector, regular and unstructured data will be covered and adequate elements of the software described to give participants the basics of VisNow usage.
- About presenter:
- Jędrzej Nowosielski graduated from Individual Inter-Faculty Studies in Mathematics and Natural Sciences at the University of Warsaw (MSc in physics). He received PhD in physics from Heriot-Watt University in Edinburgh (UK) in 2014. He currently works at Interdisciplinary Centre for Mathematical and Computational Modelling, University of Warsaw (ICM UW). He is a member of VisNow development team. His research interests include visual analysis of medical images, machine learning and particularly deep learning in the context of image data as well as explainable artificial intelligence (XAI).
Tutorial 3:
Introduction to quantum programming
- Start time: 9:00
- Presenter:
- Jaroslaw Miszczak, PhD, Institute of Theoretical and Applied Informatics, Polish Academy of Sciences
- Abstract:
- The goal of this tutorial is to provide a hands-on introduction to quantum computing and present current software solution developed for the purpose of accessing quantum computing resources. We start with the basic implementation of toy-model of a quantum computer and utilize it to demonstrate basic features of quantum computing. Next, we will utilize IBM Q platform and Qiskit package for demonstrating important operations utilized in quantum computing. We will also use this platform for implementing basic quantum algorithms.
- Important Notes:
- The tutorial will be based on Python programming language and Qiskit software package (https://qiskit.org/) developed by IBM. Installation instructions are available at https://qiskit.org/documentation/install.html
- About presenter:
- Jarosław Miszczak works at the Institute of Theoretical and Applied Informatics, Polish Academy of Sciences in Gliwice, Poland. His main research interests include quantum computing and quantum programming languages, as well as mathematical foundations of quantum mechanics. During his research work, he developed several software packages focused on the simulation of quantum computing, targeting popular scientific computing platforms.
Tutorial 4:
Introduction to scientific computing using the Julia language
- Start time: 9:00
- Presenters:
- Bogumił Kamiński, PhD, Associate Professor, SGH Warsaw School of Economics
- Przemysław Szufel, PhD, Assistant Professor, SGH Warsaw School of Economics
- Abstract:
- The first part of the tutorial is planned to be a gentle introduction to scientific computing using the Julia language. No prior knowledge of the Julia language is required, but it is assumed that the participants will have previous experience in scientific computing in general (using e.g. Python, C/C++, Matlab, …).
During the workshop we will cover:- Julia installation and environment configuration;
- Introduction to Julia syntax;
- Basics of package management in Julia;
- Solving selected example numerical computing tasks using the Julia language;
- Plotting;
- Loading and managing data (aka. data frames);
- Integration of Julia with other languages (Python, R, Matlab)
- About presenters:
- Bogumił Kamiński is the Head of Decision Analysis and Support Unit at SGH Warsaw School of Economics and Adjunct Professor at Data Science Laboratory, Ryerson University, Toronto. He is a member of the Management Committee of European Social Simulation Association (ESSA), and Vice President of Institute for Operations Research and Management Sciences (INFORMS) Polish Chapter. His field of expertise is operations research, with special focus on industrial applications of forecasting, optimization and simulation. He has been involved in development of core Julia language and its packages related to data science workflow. He is one of the top answerers for julia-lang tag on StackOverflow.
- Przemysław Szufel is an Assistant Professor in Decision Support and Analysis Unit at Warsaw School of Economics. He is a member of the Management Committee of European Social Simulation Association (ESSA). His current research focuses on methods for execution of large-scale simulations for numerical experiments and optimization. He is an author or a co-author of several Open Source tools for high performance and numerical simulation (such as KissCluster, D-MASON, Isislab SOF, SilverDecisions, PyCX), and actively participates in their development and a co-author of various algorithms for distributed simulation-optimization models (such as AKG, AOCBA).
Tutorial 5:
Parallel computing using the Julia language
- Start time: 14:00
- Presenters:
- Bogumił Kamiński, PhD, Associate Professor, SGH Warsaw School of Economics
- Przemysław Szufel, PhD, Assistant Professor, SGH Warsaw School of Economics
- Abstract:
- The second part of the tutorial is planned to show the functionalities available in Julia that make it suitable for high performance computing (HPC). It is assumed as a follow-up to introduction presented in the first part of the workshop. We plan to cover the following topics:
- Design of the Julia compiler (i.e. what makes Julia fast);
- Basics of multiple dispatch in Julia and why it matters for performance;
- Benchmarking and optimizing Julia code;
- Tricks for high performance on a single thread (SIMD, static arrays);
- Multi-threading in Julia;
- Multiprocessing and distributed computing in Julia;
- GPU computing in Julia
- About presenters:
- Bogumił Kamiński is the Head of Decision Analysis and Support Unit at SGH Warsaw School of Economics and Adjunct Professor at Data Science Laboratory, Ryerson University, Toronto. He is a member of the Management Committee of European Social Simulation Association (ESSA), and Vice President of Institute for Operations Research and Management Sciences (INFORMS) Polish Chapter. His field of expertise is operations research, with special focus on industrial applications of forecasting, optimization and simulation. He has been involved in development of core Julia language and its packages related to data science workflow. He is one of the top answerers for julia-lang tag on StackOverflow.
- Przemysław Szufel is an Assistant Professor in Decision Support and Analysis Unit at Warsaw School of Economics. He is a member of the Management Committee of European Social Simulation Association (ESSA). His current research focuses on methods for execution of large-scale simulations for numerical experiments and optimization. He is an author or a co-author of several Open Source tools for high performance and numerical simulation (such as KissCluster, D-MASON, Isislab SOF, SilverDecisions, PyCX), and actively participates in their development and a co-author of various algorithms for distributed simulation-optimization models (such as AKG, AOCBA).
Tutorial 6:
Introduction to deep neural networks with pytorch
- Duration: 4h
- Start time: 9:00
- Brief agenda:
-
- Introduction to data analysis with deep neural networks in HPC environment.
- hands-on 1: Introduction to effective training of CNNs – methods & tricks for optimizing results.
- hands-on 2: Introduction to word embeddings.
- hands-on 3: Introduction to distributed federated learning.
- Presenters:
- Wojciech Rosinski, Interdisciplinary Centre for Mathematical and Computational Modelling University of Warsaw, Poland
Lukasz Gorski, PhD, Interdisciplinary Centre for Mathematical and Computational Modelling University of Warsaw, Poland
Norbert Kapinski, PhD, Interdisciplinary Centre for Mathematical and Computational Modelling University of Warsaw, Poland - Abstract:
- Deep Neural Networks (DNNs) combined with proper infrastructure provide unique possibilities for data analysis. Thus the goal of this tutorial is to give an introduction to effective methods for utilizing DNNs in HPC. The workshop will start with a brief overview of the deep learning concept and most common neural network architectures. Further, three hands-ons will be presented aiming at image and text data processing in distributed environment. The examples and case-studies will be given with the use of PyTorch, thus basic programming skills in python are needed.
- About presenter:
- Łukasz Górski received PhD in computer science from Polish Academy of Sciences, Institute of Computer Science in 2017 and in law from Nicolaus Copernicus University in 2019. His work experience includes high performance computing in Java as well as NLP processing for legal informatics. He currently works at Interdisciplinary Centre for Mathematical and Computational Modelling.
Wojciech Rosinski is a machine learning engineer experienced in diverse R&D projects who has been actively participating in Machine Learning competitions with multiple high finishes, especially in Computer Vision domain, which earned him a Kaggle Master tier. He graduated from Warsaw University of Technology, where the topic of his MSc thesis revolved around Deep Learning in Materials Science. Currently, he works at Interdisciplinary Centre for Mathematical and Computational Modelling.
Norbert Kapinski is a PhD candidate at Polish Academy of Science. His work regards medical image processing, particularly with use of Machine Learning and Computer Vision methods. As of 2011 he’s been working on multiple R&D projects concerning medical image computing e.g. in orthopedy, biomechanics and cardiology fields. He currently works at Interdisciplinary Centre for Mathematical and Computational Modelling.
Tutorial 7:
Introduction to GPU programming using CUDA framework
- Duration: 4h
- Start time: 14:00
- Brief agenda:
-
- Single Introduction Multiple Data (SIMD) computing
- nVIDIA CUDA hardware model
- CUDA Toolchain
- Simple kernel, parallel reduction, svp
- Basic performance tuning
- External libraries: Thrust, cuBB, cuBLAS, …
- Presenters:
- Michał Dzikowski, PhD, Interdisciplinary Centre for Mathematical and Computational Modelling University of Warsaw, Poland
Grzegorz Gruszczyński, Interdisciplinary Centre for Mathematical and Computational Modelling University of Warsaw, Poland - Abstract:
- The rapid rise of tools and libraries harnessing GPUs power makes them available in a broad range of applications. While it is quite common that only top-level libraries suffice, in some cases it is required to develop the CUDA-based applications specifically for the task. This tutorial is aimed to provide an introduction to GPU programming using the CUDA toolchain. After a brief introduction into the architecture model, common low-level libraries aiding code development and optimization will be presented. The previous knowledge of C/C++ is needed to fully benefice from the tutorial.
- About presenter:
- Michal Dzikowski is one of the co-authors of the GPU based CFD code TCLB based on Lattice-Botlzmann method. His both masters and PhD thesis were based on GPU computing on CUDA framework. His research is focused on multiphase flow simulations in geological settings, especially in unsteady flow in fractured media.
Grzegorz Gruszczynski is one of the one contributors of the GPU based CFD code TCLB based on Lattice-Boltzmann method.
He works in an HPC support team at Interdisciplinary Centre for Mathematical and Computational Modelling at the University of Warsaw.
He is a PhD candidate in Computational Fluid Mechanics. His research focuses on multiphase flows and heat transfer problems.
Tutorial 8:
Introduction to the NEC SX-Aurora TSUBASA Vector Engine
Nicolas Weber, PhD, Senior Researcher, NEC Laboratories Europe
PCIe accelerator card. It has 48GB HBM2 memory with an access bandwidth
of 1.35TB/s. With only eight powerful cores using long vector units of
256 double precision words and standard programming languages like C,
C++ and Fortran the vector engine (VE) takes a different approach to
accelerators than GPGPUs, being conceptually closer to normal CPUs. This
tutorial aims at introducing the most relevant aspects of the VE:
– hardware: cores, memory, systems
– VEOS: the host offloaded operating system
– programming models: native, offloaded, reverse-offloaded, MPI hybrid
– vectorization techniques from SIMD to long vectors
– deep learning and AI with the Aurora VE
Nicolas Weber is Senior Researcher at the NEC Laboratories Europe. He received his PHD on automated memory access optimizations for GPU in 2017 from TU Darmstadt. Since then he focusses on the efficient mapping of artificial intelligence workloads onto various accelerator processors (e.g. NEC SX-Aurora or GPUs), to transparently increase performance and efficiency.