Terascale Statistics School 2023

3 Jul 2023, 00:30 → 6 Jul 2023, 20:00 Europe/Berlin

Flash hall (DESY Hamburg)

Isabell Melzer-Pellmann (CMS (CMS-Experiment)), Olaf Behnke (CMS (CMS Fachgruppe TOP))

This is a basics course, providing a round-trip through statistical methods and treatments used in data analyses in HEP

Lectures and interactive Tutorials on 

• Probability
• Hypothesis testing
• Parameter estimation
• Confidence intervals
• Systematics 
• Machine Learning
• Intro to R

given by

• Roger Barlow (Huddersfield)
• Roman Kogler (DESY)
• Harrison Prosper (Florida State University)
• Ivo van Vulpen (Nikhef)

It is expected that you can bring your own laptop for the excercises. 

The school fee is 80 Euro.

Please register until 18 June 2023.
 

For the Q&A session on thursday, July 6 at 2pm, please dump your questions in this editable google docs file  

 

2023-07-Statistics_School-Poster.pdf

map_terascale_stat_23.pdf

Monday, 3 July

10:00 → 10:30 Registration

10:30 → 10:45 Welcome

Speakers: Dr Isabell Melzer-Pellmann (CMS (CMS-Experiment)), Olaf Behnke (CMS (CMS Fachgruppe TOP))

obehnke230603.pdf

10:45 → 11:30 Basics Part I

Probability, Frequentist and Bayesian. Confidence. Bayes theorem. Priors
and posteriors

Speaker: Roger Barlow (Huddersfield)

Lecture1.pdf

11:30 → 12:00 Discussion Time

12:00 → 13:30 Lunch Break

13:30 → 14:15 Basics Part II

Probability distributions (Binomial and Poisson) and Probability
distribution functions (Gaussian). Expectation values. Hypothesis testing

Speaker: Roger Barlow (Huddersfield)

Lecture2.pdf

14:15 → 14:30 Discussion time

14:30 → 15:00 Coffee break

15:00 → 18:00 Higgs Analysis Walk Through Tutorial Part 1

DESY Terascale Statistics School - July 2023

What: description analysis walkthrough session

Who: Oliver Rieger, Zef Wolffs, Ivo van Vulpen

Course description

The analysis walkthrough is a hands-on session in which we will address, in a ‘real life’ example, several of the statistics topics that were covered in the lectures. More concretely: we will study the four-muon invariant mass distribution in the Higgs boson decay to four charged muons to explore three different topics:


1. Significance (Poisson distribution, p-values, optimizations)

2. Likelihood fits (parameter estimation, side band fits, background uncertainty)
3. Hypothesis testing (test statistic, toy data-sets, limits)

Goal of the exercises is to guide participants through the various steps without using the standard toolkits, i.e. we’ll focus on the concepts and program as much as we can ourselves.

Set-up and computing requirements:
After an introduction lecture all exercises and related background information can be accessed through a dedicated website.
As participant you can use the Root set-up on your laptop (Option 1 below),
or the DESY computing cluster (Option 2 below) for which guest accounts will be provided by the workshop organisers.
Through a GitLab repository all participants have access to all the material: exercises, data-sets, scripts, background information and, not unimportant, the answers to the exercises. Details on the GitLab repository and the website will be provided later.

---

Option 1: Run Root locally on the laptop

We will run the software locally on the laptop. This requires that you have
Root installed. If that is not the case you can follow the instructions
here: https://root.cern/install/

---

Option 2: Run on DESY cluster from laptop

We have rented guest accounts on the DESY Cluster, where
tutorials can be run. The only requirement for participants is a
laptop that allows an SSH connection.
- For Linux and MacOS this can be done directly from the terminal
- For Windows machines you need an SSH client (Putty) or chrome extension
secure SSH or setup via VS code and SSH keys & config.

ssh -Y schoolxx@naf-cms.desy.de with xx in range [00,79]
pwd: p9FqFt7f

Speaker: Ivo van Vulpen (NIKHEF)

2023_07_03_Statisticsschool_DESY.pdf

2023_07_03_Statisticsschool_DESY_with_solutions.pdf

Tuesday, 4 July

09:30 → 10:15 Basics Part III

Basics Estimation. Maximum likelihood. Least squares. Fitting histograms. Chi squared and goodness of fit. p-value

Speaker: Roger Barlow (Huddersfield)

Lecture3.pdf

10:15 → 10:45 Discussion time

10:45 → 11:15 Coffee break

11:15 → 12:00 Confidence Interval Estimation Part I

Intricately linked with the estimation of parameters is the question on how
to obtain meaningful uncertainties on the obtained parameters. In these two
lectures, we will look at this problem in detail and discuss confidence
intervals on extracted parameters. We will start with a simple case of a
counting experiment following Poissonian statistics. This will lead us to
the coverage of uncertainties, which we will use to study different
estimators for the statistical uncertainty. We will continue with the more
general case of how to obtain confidence intervals for a true parameter,
given a measurement of a quantity related to this parameter. The lecture
will cover confidence intervals close to physical boundaries and limit
setting with the CLs method.

Speaker: Roman Kogler (DESY FH, CMS)

Confidence_Intervals_part1.pdf

exercise_coverage.pdf

12:00 → 12:30 Discussion time

12:30 → 14:00 Lunch break

14:00 → 14:45 Confidence Interval Estimation Part II

Speaker: Roman Kogler (DESY FH, CMS)

Confidence_Intervals_part2.pdf

14:45 → 15:00 Discussion time

15:00 → 15:30 Coffee Break

15:30 → 18:30 Higgs Analysis Walk Through Tutorial Part 2

Speaker: Ivo van Vulpen (NIKHEF)

19:00 → 21:00 Workshop Dinner

Canteen Extension, DESY Hamburg

Wednesday, 5 July

09:30 → 10:15 Systematics Part 1

Estimation and checks

Speaker: Roger Barlow (Huddersfield)

Lecture4.pdf

10:15 → 10:45 Discussion time

10:45 → 11:15 Coffee break

11:15 → 12:00 Machine Learning Part 1

Title: An Introduction to Machine Learning

Abstract:

The three lecture/tutorials cover the basic principles of machine learning,
which is the underlying technology of artificial intelligence. Lecture 1
covers the mathematical foundations with a focus on supervised learning.
Lectures 2 and 3 cover a few of the widely used machine learning models,
including boosted decision trees, deep feed forward neural networks,
convolutional neural networks, auto-encoders and if time permits,
transformers (the model underlying ChatGPT). The models will be illustrated
with simples example from particle physics, astronomy, and calculus.

Link to the software for the Computer Tutorials:
https://github.com/hbprosper/Terascale

Speaker: Harrison Prosper (FSU)

Terascale2023_Prosper-1.pdf

12:00 → 12:30 Discussion time

12:30 → 14:00 Lunch break

14:00 → 14:45 Systematics Part II

Correlation, building and handling the matrices

Speaker: Roger Barlow (Huddersfield)

Lecture5.pdf

14:45 → 15:00 Discussion time

15:00 → 15:30 Coffee break

15:30 → 18:00 Machine Learning Part 2

Speaker: Harrison Prosper (FSU)

Terascale2023_Prosper-2.pdf

Thursday, 6 July

09:30 → 10:15 Machine Learning Part 3

Speaker: Harrison Prosper (FSU)

Terascale2023_Prosper-3.pdf

10:15 → 10:45 Discussion time

10:45 → 11:15 Coffee break

11:15 → 12:00 Introduction to R

An introduction to the R language. The aim of this is to explain the
minimum about R that everybody needs to know, and hopefully encourage those
that would benefit from learning and using the language to do so.

For Lecture 6, it will be helpful (though not essential) to download R
beforehand from https://cran.r-project.org/ [cran.r-project.org]
or, if you prefer working in an IDE, R-studio from
https://posit.co/download/rstudio-desktop/ [posit.co]

Speaker: Roger Barlow (Huddersfield)

Lecture6.pdf

12:00 → 12:30 Discussion time

12:30 → 14:00 Lunch break

14:00 → 14:45 Special Q&A session

We will discuss questions collected at:
https://docs.google.com/document/d/1id8UczLtlOWlhI7LMXK8cU37AE_wAbIQzyo1nodgxWI/edit

14:45 → 15:00 Discussion time

15:00 → 15:45 Artificial Intelligence Today and Tomorrow

Recent advances in the field of artificial intelligence (AI) have given us a
glimpse of a potentially thrilling, even civilization-changing, future.
From that optimistic viewpoint, after a brief historical introduction, I
survey some of the recent advances in AI. Then, I speculate about what the
impact of these advances might be on the nature of research in particle
physics over the next few decades. I end by acknowledging the fact that
every technology, however benign it may at first appear, can be used for
good or for ill. It is, therefore, appropriate to sound a cautionary note,
not so much to be a doomsayer, which I'm not, but rather to remind you that
the future remains yours to make.

Speaker: Harrison Prosper

15:45 → 16:00 Discussion time

16:00 → 16:10 Closing/Good bye

Speakers: Isabell Melzer-Pellmann (CMS (CMS-Experiment)), Olaf Behnke (CMS (CMS Fachgruppe TOP))

obehnke230603_final.pdf

16:10 → 16:40 Coffee break