Inverted CERN School of Computing 2019

4 Mar 2019, 08:30 → 7 Mar 2019, 11:00 Europe/Zurich

31/3-004 - IT Amphitheatre (CERN)

The 12th Inverted CERN School of Computing (iCSC 2019) consists of lectures and hands-on exercises presented over a few days by former CERN School of Computing students. The Inverted School provides a platform to share their knowledge by turning the students into teachers. More information on the Inverted CSC events can be found at http://csc.web.cern.ch/inverted-school.

Topics covered this year include:

Artificial Intelligence Machine Learning Pattern Recognition

Big Data Container Orchestration Tensor Networks

Computational Physics Numerical Analysis Track Finding

Keras

TensorFlow

FPGAs

Kubernetes

Hadoop

Spark

The event will take place on March 4-7, 2019 at CERN: in the IT Amphitheatre / 31-3-004 (lectures) and in 513-1-024 (exercises - please bring your laptop). The attendance is free, and open to anyone at CERN. We also webcast the lectures (but not exercises) so that you can tune in from anywhere in the world.

Register, and you can ask for the iCSC 2019 booklet (distributed during the event, while stocks last!). We'll also make sure enough tea/coffee is available if we know you're coming.

                                                             

iCSC 2019 Booklet.pdf

iCSC 2019 Poster.pdf

Monday, 4 March

08:30 → 09:00 Welcome coffee

09:00 → 09:10 A word from the IT Department Head

Speaker: Frederic Hemmer (CERN)

Recording

09:10 → 09:15 Introduction to the Inverted CSC

Speaker: Sebastian Lopienski (CERN)

2019.03.04 iCSC-2019 Welcome.pdf

2019.03.04 iCSC-2019 Welcome.pptx

Recording

09:15 → 10:15 A practical approach to Convolutional Neural Networks (lecture)

The field of Artificial Intelligence is blooming with many techniques and developments in the last years. Convolutional Neural Networks is a class of feed-forward neural networks that are typically used to analyze images and extract information.

We will explore the mathematical foundation of CNNs, join the hype and get some practical hands-on experience with the technology that will change our everyday lives.

Speaker: Daniel Hugo Campora Perez (CERN + University of Seville)

Daniel Campora - A practical approach to Convolutional Neural Networks_slides.pdf

Recording

10:15 → 10:45 Coffee

10:45 → 11:45 A Scientist's Guide to FPGAs

Field Programmable Gate Arrays (FPGAs) have become ubiquitous in a variety of technological and scientific fields. Their versatility make them an ideal match not only for computing intensive tasks but also for the differing requirements of custom electronics that often can be found in experimental setups.

This seminar leads the audience into the fully programmable and intrinsically parallel world of FPGAs. After an introduction to digital design and the anatomy of an FPGA, the design flow and required way of thinking will be presented. The seminar will be completed by a comparison of hardware and software-driven computation as well as an overview of the application of FPGAs in different fields and tasks.

Speaker: Alexander Ruede (CERN + KIT-IPE)

Alexander Ruede - A Scientist's Guide to FPGAs_AScientistsGuideToFPGAs.pdf

Recording

11:45 → 13:30 Lunch break / WIT Diversity Talk

During lunchtime, CERN's Women in Technology (WIT) community will have a Diversity Talk with Tim Smith. More details: https://indico.cern.ch/event/792031/

Inverted CSC participants are welcome to join this very interesting event.

WITintro_iCSC.pdf

WITintro_iCSC.pptx

13:30 → 14:30 Global track finding algorithms

In high energy physics experiments the reconstruction of tracks of charged particles provides the core for the measurements of these particles' properties. Track finding algorithms can roughly be divided into two main categories: local and global. Local track finding algorithms try to link individual hits one by one while utilizing a variety of smart techniques to mitigate combinatorial complexity, whereas global track finding algorithms treat all hits simultaneously.

In this lecture we will look into track finding algorithms in wire chambers that are performed on all hits at once. The search is done by converting individual hit parameters to a curve in a dual space using Legendre or Hough transform, with the intersection of multiple curves corresponding to a track compatible with given hits. Then, the problem of finding a track is translated to the problem of finding most densely populated regions in the dual space which can be done effectively and quickly by a quadtree search.

Speaker: Dmitry Neverov (Nagoya University)

Dmitry Neverov - Global track finding algorithms_Global_track_finding_v1.0.pdf

Global_track_finding_v1.1.pdf

Recording

14:30 → 15:00 Coffee

15:00 → 16:00 Hardware Acceleration Through FPGAs - Basic Concepts (lecture 1)

FPGAs are a more and more ubiquitous technology. They offer the benefits of fast, application-tailored hardware, typically associated with ASICs, while enabling fast prototyping, upgradability and low costs. This makes them an ideal ally in HEP computing, specifically in areas where high performance is needed and/or specifications and needs may vary.

The lectures will focus on the intrinsic parallel processing characteristics of FPGAs, emphasizing how they can be exploited to implement data-intensive algorithms. Focus will also be put on concepts like hardware/software partitioning (very important to help the most performing parts of the systems in collaborating with the legacy CPU oriented codebase).

A simple hands on exercises session will be added to let the students get acquainted with the main tools and the VHDL language.

Speaker: Mr Giorgio Lopez (CERN)

Giorgio Lopez - Hardware Acceleration Through FPGAs - Basic Concepts_iCSC lecture 1-icsc1-final.pdf

Recording

16:00 → 17:00 A practical approach to Convolutional Neural Networks (exercise 1)

The field of Artificial Intelligence is blooming with many techniques and developments in the last years. Convolutional Neural Networks is a class of feed-forward neural networks that are typically used to analyze images and extract information.

We will explore the mathematical foundation of CNNs, join the hype and get some practical hands-on experience with the technology that will change our everyday lives.

Speaker: Daniel Hugo Campora Perez (CERN + University of Seville)

1. Log in to CERN box

2. Log in into SWAN

3. From SWAN, click on "Download Project from git"

17:00 → 18:00 A practical approach to Convolutional Neural Networks (exercise 2)

The field of Artificial Intelligence is blooming with many techniques and developments in the last years. Convolutional Neural Networks is a class of feed-forward neural networks that are typically used to analyze images and extract information.

We will explore the mathematical foundation of CNNs, join the hype and get some practical hands-on experience with the technology that will change our everyday lives.

Speaker: Daniel Hugo Campora Perez (CERN + University of Seville)

Tuesday, 5 March

08:30 → 09:00 Welcome coffee

09:00 → 10:00 Hardware Acceleration Through FPGAs - Basics of VHDL (lecture 2)

FPGAs are a more and more ubiquitous technology. They offer the benefits of fast, application-tailored hardware, typically associated with ASICs, while enabling fast prototyping, upgradability and low costs. This makes them an ideal ally in HEP computing, specifically in areas where high performance is needed and/or specifications and needs may vary.

The lectures will focus on the intrinsic parallel processing characteristics of FPGAs, emphasizing how they can be exploited to implement data-intensive algorithms. Focus will also be put on concepts like hardware/software partitioning (very important to help the most performing parts of the systems in collaborating with the legacy CPU oriented codebase).

A simple hands on exercises session will be added to let the students get acquainted with the main tools and the VHDL language.

Speaker: Mr Giorgio Lopez (CERN)

Giorgio Lopez - Hardware Acceleration Through FPGAs - Basics of VHDL_iCSC lecture 2- icsc2-final.pdf

Recording

10:00 → 11:00 Tensor Networks - Introduction and Matrix Product States (lecture 1)

In recent years, tensor networks have become a viable alternative to Monte Carlo calculations and exact diagonalization for the simulation of many-body systems.
As they represent a formulation of quantum mechanical wavefunctions with polynomially many parameters, they make calculations of large systems feasible.

They have already found wide application in condensed matter physics and start to be an interesting tool for high energy physics as well.

In this lecture series, I will introduce the basic concepts of tensor networks.

We will start with an introduction of the necessary basics of quantum mechanics and linear algebra and focus on the algorithmic side of tensor networks in the second lecture.

Speaker: Patrick Emonts (Max Planck Institute of Quantum Optics)

Recording

Tensor_Networks_handout.pdf

Tensor_Networks_lecture.pdf

11:00 → 11:15 Coffee

11:15 → 12:15 Tensor Networks - The iTEBD Algorithm (lecture 2)

In recent years, tensor networks have become a viable alternative to Monte Carlo calculations and exact diagonalization for the simulation of many-body systems.
As they represent a formulation of quantum mechanical wavefunctions with polynomially many parameters, they make calculations of large systems feasible.

They have already found wide application in condensed matter physics and start to be an interesting tool for high energy physics as well.

In this lecture series, I will introduce the basic concepts of tensor networks.

We will start with an introduction of the necessary basics of quantum mechanics and linear algebra and focus on the algorithmic side of tensor networks in the second lecture.

Speaker: Patrick Emonts (Max Planck Institute of Quantum Optics)

Recording

Tensor_Networks_handout.pdf

Tensor_Networks_lecture.pdf

12:15 → 13:30 Lunch break

13:30 → 14:30 Big Data Technologies and Physics Analysis with Apache Spark (lecture 1)

The Large Hadron Collider is scheduled to shut down for a 2 years maintenance period since December 2018. However, the already collected data -which are stored in a dedicated custom storage service- between April 2015 and November 2018, exceed 150 PBs in total. To analyse these data, more and more teams at CERN decide to use Big Data Technologies to perform Physics Analysis and "Data Reduction", i.e. produce smaller reusable datasets for frequent access. These technologies show great potential in speeding up the existing procedures.

This lecture will provide an overview of the latest trending big data technologies in the Hadoop and Spark ecosystems with focus on their main architecture characteristics, and then will target a number of important questions: How can we perform Physics Analysis with Big Data Technologies? What are the problems faced? What are the challenges and the available data sources? What are the other domain in which Big Data Analytics are applied at CERN?

Speaker: Evangelos Motesnitsalis (CERN)

Evangelos Motesnitsalis - Big Data Technologies and Physics Analysis with Apache Spark_iCSC Slides - final version.pdf

Recording

14:30 → 15:30 Big Data Technologies and Physics Analysis with Apache Spark (lecture 2)

The Large Hadron Collider is scheduled to shut down for a 2 years maintenance period since December 2018. However, the already collected data -which are stored in a dedicated custom storage service- between April 2015 and November 2018, exceed 150 PBs in total. To analyse these data, more and more teams at CERN decide to use Big Data Technologies to perform Physics Analysis and "Data Reduction", i.e. produce smaller reusable datasets for frequent access. These technologies show great potential in speeding up the existing procedures.

This lecture will provide an overview of the latest trending big data technologies in the Hadoop and Spark ecosystems with focus on their main architecture characteristics, and then will target a number of important questions: How can we perform Physics Analysis with Big Data Technologies? What are the problems faced? What are the challenges and the available data sources? What are the other domain in which Big Data Analytics are applied at CERN?

Speaker: Evangelos Motesnitsalis (CERN)

Evangelos Motesnitsalis - Big Data Technologies and Physics Analysis with Apache Spark_iCSC Slides - final version.pdf

Recording

15:30 → 16:00 Coffee

16:00 → 17:00 Hardware Acceleration Through FPGAs - First Experiments in VHDL (exercise 1)

FPGAs are a more and more ubiquitous technology. They offer the benefits of fast, application-tailored hardware, typically associated with ASICs, while enabling fast prototyping, upgradability and low costs. This makes them an ideal ally in HEP computing, specifically in areas where high performance is needed and/or specifications and needs may vary.

The lectures will focus on the intrinsic parallel processing characteristics of FPGAs, emphasizing how they can be exploited to implement data-intensive algorithms. Focus will also be put on concepts like hardware/software partitioning (very important to help the most performing parts of the systems in collaborating with the legacy CPU oriented codebase).

A simple hands on exercises session will be added to let the students get acquainted with the main tools and the VHDL language.

Speaker: Mr Giorgio Lopez (CERN)

Exercises.docx

Exercises.pdf

exercises setup instructions PLEASE DO IN ADVANCE.rtf

Solutions.zip

17:00 → 18:00 Hardware Acceleration Through FPGAs - Easy DSP Applications (exercise 2)

FPGAs are a more and more ubiquitous technology. They offer the benefits of fast, application-tailored hardware, typically associated with ASICs, while enabling fast prototyping, upgradability and low costs. This makes them an ideal ally in HEP computing, specifically in areas where high performance is needed and/or specifications and needs may vary.

The lectures will focus on the intrinsic parallel processing characteristics of FPGAs, emphasizing how they can be exploited to implement data-intensive algorithms. Focus will also be put on concepts like hardware/software partitioning (very important to help the most performing parts of the systems in collaborating with the legacy CPU oriented codebase).

A simple hands on exercises session will be added to let the students get acquainted with the main tools and the VHDL language.

Speaker: Mr Giorgio Lopez (CERN)

Wednesday, 6 March

08:30 → 09:00 Welcome coffee

09:00 → 10:00 Tensor Networks - Singular Value Decomposition (exercise 1)

In recent years, tensor networks have become a viable alternative to Monte Carlo calculations and exact diagonalization for the simulation of many-body systems.
As they represent a formulation of quantum mechanical wavefunctions with polynomially many parameters, they make calculations of large systems feasible.

They have already found wide application in condensed matter physics and start to be an interesting tool for high energy physics as well.

In this lecture series, I will introduce the basic concepts of tensor networks.

We will start with an introduction of the necessary basics of quantum mechanics and linear algebra and focus on the algorithmic side of tensor networks in the second lecture.

Speaker: Patrick Emonts (Max Planck Institute of Quantum Optics)

setup_instructions.pdf

10:00 → 11:00 Tensor Networks - Application of the iTEBD Algorithm (exercise 2)

In recent years, tensor networks have become a viable alternative to Monte Carlo calculations and exact diagonalization for the simulation of many-body systems.
As they represent a formulation of quantum mechanical wavefunctions with polynomially many parameters, they make calculations of large systems feasible.

They have already found wide application in condensed matter physics and start to be an interesting tool for high energy physics as well.

In this lecture series, I will introduce the basic concepts of tensor networks.

We will start with an introduction of the necessary basics of quantum mechanics and linear algebra and focus on the algorithmic side of tensor networks in the second lecture.

Speaker: Patrick Emonts (Max Planck Institute of Quantum Optics)

setup_instructions.pdf

11:00 → 11:15 Coffee

11:15 → 12:15 How container orchestration can strengthen your micro-services: the approach of Kubernetes (lecture)

With the rise of container technologies during the past few years there have been many paradigms shifts in terms of software development, deployment and maintenance, especially in conjunction with micro-service architectures.

The lecture covers these fundamental concepts and focuses on the challenges of container orchestration. Crucial aspects like horizontal and vertical scaling, availability, fault-tolerance and rolling updates are among the topics covered by the lecture which will then also be experienced during the hands-on exercises.

Speaker: Riccardo Poggi (University of Geneva)

GitLab Exercise Repository

Recording

Riccardo_Poggi_microservices_docker_and_kubernetes.pdf

12:15 → 13:00 Lunch break

13:00 → 14:30 Data Centre visit

14:30 → 15:30 Efficient C++ implementation of custom FEM kernel with Eigen

We will start from examples of problems solved by finite element method - equilibrium magnetic fields, structural deflection calculations. Then we will talk about foundation of FEM method key concepts such as stiffness matrix and impact of high matrix dimensions and sparse characteristic to ways data can be calculated more efficiently.

To implement kernel we will introduce Eigen, a C++ linear algebra library that eliminates intermediate temporary objects by utilizing expression templates technique and generates efficient high-level math code with most of complexity taken from you.

Speaker: Mikhail Sizov (Budker Institute of Nuclear Physics)

Mikhail Sizov - Efficient C++ implementation of custom FEM kernel with Eigen_FEM_Sizov.pdf

Recording

15:30 → 16:00 Coffee

16:00 → 17:00 How container orchestration can strengthen your micro-services: the approach of Kubernetes (exercise 1)

With the rise of container technologies during the past few years there have been many paradigms shifts in terms of software development, deployment and maintenance, especially in conjunction with micro-service architectures.

The lecture covers these fundamental concepts and focuses
on the challenges of container orchestration. Crucial aspects like horizontal and vertical scaling, availability, fault-tolerance and rolling updates are among the topics covered by the lecture which will then also be experienced during the hands-on exercises.

Speaker: Riccardo Poggi (University of Geneva)

GitLab Exercise Repository

Riccardo_Poggi_microservices_docker_and_kubernetes.pdf

17:00 → 18:00 How container orchestration can strengthen your micro-services: the approach of Kubernetes (exercise 2)

With the rise of container technologies during the past few years there have been many paradigms shifts in terms of software development, deployment and maintenance, especially in conjunction with micro-service architectures.

The lecture covers these fundamental concepts and focuses
on the challenges of container orchestration. Crucial aspects like horizontal and vertical scaling, availability, fault-tolerance and rolling updates are among the topics covered by the lecture which will then also be experienced during the hands-on exercises.

Speaker: Riccardo Poggi (University of Geneva)

Thursday, 7 March

08:30 → 09:00 Welcome coffee

09:00 → 10:00 Big Data Technologies and Physics Analysis with Apache Spark (exercise 1)

The Large Hadron Collider is scheduled to shut down for a 2 years maintenance period since December 2018. However, the already collected data -which are stored in a dedicated custom storage service- between April 2015 and November 2018, exceed 150 PBs in total. To analyse these data, more and more teams at CERN decide to use Big Data Technologies to perform Physics Analysis and "Data Reduction", i.e. produce smaller reusable datasets for frequent access. These technologies show great potential in speeding up the existing procedures.

This lecture will provide an overview of the latest trending big data technologies in the Hadoop and Spark ecosystems with focus on their main architecture characteristics, and then will target a number of important questions: How can we perform Physics Analysis with Big Data Technologies? What are the problems faced? What are the challenges and the available data sources? What are the other domain in which Big Data Analytics are applied at CERN?

Speaker: Evangelos Motesnitsalis (CERN)

Big Data Technologies and Physics Analysis with Apache Spark - iCSC Exercises.pdf

10:00 → 11:00 Big Data Technologies and Physics Analysis with Apache Spark (exercise 2)

The Large Hadron Collider is scheduled to shut down for a 2 years maintenance period since December 2018. However, the already collected data -which are stored in a dedicated custom storage service- between April 2015 and November 2018, exceed 150 PBs in total. To analyse these data, more and more teams at CERN decide to use Big Data Technologies to perform Physics Analysis and "Data Reduction", i.e. produce smaller reusable datasets for frequent access. These technologies show great potential in speeding up the existing procedures.

This lecture will provide an overview of the latest trending big data technologies in the Hadoop and Spark ecosystems with focus on their main architecture characteristics, and then will target a number of important questions: How can we perform Physics Analysis with Big Data Technologies? What are the problems faced? What are the challenges and the available data sources? What are the other domain in which Big Data Analytics are applied at CERN?

Speaker: Evangelos Motesnitsalis (CERN)