Terascale Statistics School 2024

2 Apr 2024, 10:00 → 5 Apr 2024, 14:00 Europe/Berlin

Seminarraum Flash (DESY Hamburg)

Andreas Hinzmann (CMS (CMS Fachgruppe Searches)), Dirk Kruecker (CMS (CMS Fachgruppe Searches)), Isabell Melzer-Pellmann (Deutsches Elektronen-Synchrotron DESY), Olaf Behnke (CMS (CMS Fachgruppe TOP))

This is a course on statistical methods in particle physics mainly targeting PhD students but it is open to other interested students and post-docs. The school covers key analysis tasks in theory and practice as well as going deep with lectures into some specific areas of high relevance. 

It is expected that you bring your own laptop for the hands on sessions.
The school fee is 80 Euro. 

Find here a DESY map with the relevant school places

Organizing Committee:
I. Henning, A. Hinzmann, D. Kruecker, I. Melzer-Pellmann, O. Behnke

2024-04-Statistics-School-Poster.pdf

map_terascale_stat_23.pdf

program_in_a_nutshell.png

Tuesday 2 April

10:00 → 11:00 Registration

11:00 → 11:05 Welcome

Speakers: Isabell Melzer-Pellmann (Deutsches Elektronen-Synchrotron DESY), Olaf Behnke (CMS (CMS Fachgruppe TOP))

obehnke240402.pdf

11:05 → 12:30 Statistical Inference Lecture 1

Plan for Lectures 1 and 2:
- Quick review of probability, frequentist vs. Bayesian approaches
- Parameter estimation, maximum likelihood, asymptotic properties of MLEs
- Hypothesis tests – general formalism, p-values
- Confidence intervals from inversion of test, from likelihood function
- General analysis including nuisance parameters, asymptotics

Speaker: Glen Cowan (RHUL)

cowan_desy24_1.pdf

12:30 → 14:00 Lunch Break

14:00 → 15:30 Combine Tool Tutorial I

Plan for Tutorial parts I - IV:
In this tutorial we will go through the main features of the Combine software package which provides a command-line interface to many different statistical techniques, available inside RooFit/RooStats, that are used widely inside CMS.
In the first part of the tutorial we’ll explain the format of the Combine configuration file (datacard), and then discuss how to set up a simple counting experiment, followed by the shape analysis setup using ROOT histograms as inputs. Then we’ll look at how to build simultaneous fits using independent categories, add systematic uncertainties and extract the limits or uncertainty estimates for the parameters of interest. Second part of the tutorial describes how to construct parametric models with RooFit and extract the measurements using Combine.

Speakers: Aliya Nigamova (University of Hamburg), Kyle Cormier (Uni Zurich), Nicholas Wardle

Combine setup

Combine_TerascaleScholl2024.pdf

CombineTutorial_TSS_2024_Intro.pdf

15:30 → 16:00 Coffee Break

16:00 → 17:30 Combine Tool Tutorial II

Speakers: Aliya Nigamova (University of Hamburg), Kyle Cormier (Uni Zurich), Nicholas Wardle

Combine_TerascaleScholl2024.pdf

CombineTutorial_TSS_2024.pdf

Wednesday 3 April

09:15 → 10:45 Statistics for Machine Learning Lecture I

Plan for Lectures I - III:
LHC Physics with its vast amount of data and its precise first-principle simulations is a perfect case for using modern machine learning techniques. This includes data acquisition, analysis, simulations, and inference. I will introduce the main ML-concepts and ML-tools in a physics-specific manner, with a focus on statistics aspects. I will start by introducing Bayesian neural networks and likelihood loss functions and show how we can train classification networks with a probabilistic output. I will then introduce a range of generative networks, which are transforming not only our daily lives but also LHC physics. Finally, I will discuss how these ML-methods enable unfolding or inverse simulations and optimal inference for the LHC and more broadly. While there will be no tutorial for this lecture, all students are encouraged to deepen their understanding using the tutorials which are provided with the lecture notes https://arxiv.org/abs/2211.01421 Updated lecture notes are here:
https://www.thphys.uni-heidelberg.de/~plehn/pics/modern_ml.pdf

Speaker: Tilman Plehn (Heidelberg University)

10:45 → 11:15 Coffee Break

11:15 → 12:45 Statistical Inference Lecture II

Speaker: Glen Cowan (RHUL)

cowan_desy24_2.pdf

12:45 → 14:15 Lunch Break

14:15 → 15:30 Combine Tool Tutorial III

Speakers: Aliya Nigamova (University of Hamburg), Kyle Cormier (Uni Zurich), Nicholas Wardle

combine_parametric_exercise-2.pdf

CombineTutorial_TSS_2024-2.pdf

15:30 → 16:00 Coffee Break

16:00 → 17:30 Combine Tool Tutorial IV

Speakers: Aliya Nigamova (University of Hamburg), Kyle Cormier (Uni Zurich), Nicholas Wardle

Combine_TerascaleScholl2024.pdf

18:30 → 20:30 School Dinner

Thursday 4 April

09:00 → 10:30 Statistics for Machine Learning II

Speaker: Tilman Plehn (Heidelberg University)

10:45 → 11:15 Coffee Break

11:15 → 12:45 Unfolding Lecture

In high-energy physics, unfolding is a critical statistical process for interpreting experimental data that is complicated by the intrinsic ill-posedness of the problem. This complexity arises from the need to provide heuristics for statistical estimates that disentangle true physical phenomena from observational distortions. We present a typical roadmap for why, when, and how unfolding is applied in high energy physics experiments and how the treatment of uncertainties influences considerations such as the choice of algorithm and regularisation.

Speaker: Vincent Croft

Terrascale24-unfolding.pdf

12:45 → 14:00 Lunch Break

14:00 → 15:30 Special Session on Uncertainties: Introductory Lecture and Lecture on Treatment of Asymmetric Uncertainties

What you mean by ‘the error on the result’, which is not as simple as you think.
- Bayesian and frequentist probability and confidence.
- The Neyman construction.
- Statistical and systematic uncertainties
- Asymmetric uncertainties and how to handle them

Speaker: Roger Barlow (Huddersfield)

Uncertainties.pdf

15:30 → 16:00 Coffee Break

16:00 → 17:30 Special Session on Uncertainties: Lecture on Errors on Errors

A statistical model is presented where the uncertainty in assignment of systematic errors is taken into account. Estimates of the systematic variances are modeled as gamma distributed variables. The resulting confidence intervals show interesting and useful properties, e.g., when averaging measurements to estimate their mean, the size of the confidence interval increases as a for decreasing goodness-of-fit, and averages have reduced sensitivity to outliers. The basic properties of the model are presented and several types of examples relevant for Particle Physics are explored. (Based on G. Cowan, Eur. Phys. J. C (2019) 79:133, E. Canonero et al., Eur. Phys. J. C (2023) 83:1100.)

Speaker: Glen Cowan (RHUL)

cowan_desy24_3.pdf

Friday 5 April

09:15 → 10:45 Statistics of Machine Learning Lecture III

Speaker: Tilman Plehn (Heidelberg University)

10:45 → 11:15 Coffee Break + School Foto

11:15 → 12:15 School Summary

Looking back, a reminder of what you’ve learned in the week and, looking forward, how you’re likely to use it.

Speaker: Roger Barlow (Huddersfield)

Summary.pdf

12:15 → 12:30 Closing of School - Good Bye

Speakers: Isabell Melzer-Pellmann (Deutsches Elektronen-Synchrotron DESY), Olaf Behnke (CMS (CMS Fachgruppe TOP))

obehnke240402.pdf

12:30 → 14:00 Lunch Break