MLHEP 2015 summer school

27 Aug 2015, 08:00 → 30 Aug 2015, 23:00 Europe/Moscow

Academic University

Summer school on Machine Learning in High Energy Physics http://www.hse.ru/mlhep2015/

Kaggle COMET competition

MLHEP github repo

Thursday 27 August

Registration

08:15 breakfast

Welcome

Convener: Andrey Ustyuzhanin (Yandex School of Data Analysis (RU))

1 Welcome to MLHEP

2015-08_MLHEP_Welcome.pdf

2 MLHEP Kaggle competition

Speaker: Nikita Kazeev (Yandex School of Data Analysis (RU))

mlhep2015_kaggle_slides.pdf

09:35 break

Advanced track: lectures

Convener: Dr Victor Kitov (MSU)

3 Reminder about major algorithms. Advanced aspects of their use. Model evaluation.

Introduction. Model evaluation.pdf

11:00 coffee break

4 Feature selection. Regularization.

Feature selection.pdf

Introductory track: lectures

Convener: Alexey Rogozhnikov

5 Intro: General pipeline, ML at a glance

lecture1.pdf

11:00 coffee break

6 Intro: kNN, basic overfitting, roc curve, logistic regression

12:40 lunch

Advanced track: seminars

Convener: Tatiana Likhomanenko (National Research Centre Kurchatov Institute (RU))

7 MLHEP introduction. ML vs MLHEP: correlation and agreement restrictions

Speaker: Tatiana Likhomanenko (National Research Centre Kurchatov Institute (RU))

seminar1_part1.pdf

15:15 coffee break

8 sPlot technique. ML on sPlot data.

seminar1_part2.pdf

Introductory track: seminars

Convener: Nikita Kazeev (Yandex School of Data Analysis (RU))

9 Introduction, course technicalities

1.1 Technicalities and principles.ipynb

1.2 Tools numpy pandas matplotlib - seminar.ipynb

15:15 coffee break

10 NumPy, Pandas, Matplotlib

17:10 break

11 Applying Machine Learning to LHC triggers optimization

Speaker: Mika Anton Vesterinen (Ruprecht-Karls-Universitaet Heidelberg (DE))

Mika_MLHEP.pdf

19:00 Welcome dinner

Friday 28 August

08:15 breakfast

Advanced track: lectures #2

Convener: Dr Victor Kitov (MSU)

12 model ensembling #1

10:50 coffee break

13 model ensembling #2

Introductory track: lectures #2

Convener: Alexey Rogozhnikov

14 neural networks

lecture2.pdf

10:50 coffee break

15 decision tree, regression tree

12:30 lunch

Advanced track: seminars #2

Convener: Tatiana Likhomanenko (National Research Centre Kurchatov Institute (RU))

16 Brave new boosting world: flatness boosting

seminar2.pdf

15:15 coffee break

17 Brave new boosting world: reweighing boosting. Iterative learning

Introductory track: seminars #2

Convener: Nikita Kazeev (Yandex School of Data Analysis (RU))

18 Overfitting, Cross-validation

2.1. Basic ML, validation, feature engineering.ipynb

15:15 coffee break

19 Sklearn: KNN, decision tree, logistic regression

17:10 break

20 Challenges of searching for physics signatures with a large amount of background

Speaker: Patrick Haworth Owen (Imperial College Sci., Tech. & Med. (GB))

Patrick_Machine_Learning_in_LHCb.pdf

Saturday 29 August

08:15 breakfast

Advanced track: lectures #3

Convener: Dr Victor Kitov (MSU)

21 Linear dimensionality reduction.

Dimensionality reduction.pdf

10:50 coffee break

22 Non-linear dimensionality reduction. Kernel trick. Common kernels.

Introductory track: lectures #3

Convener: Alexey Rogozhnikov

23 Ensembles

lecture3.pdf

10:50 coffee break

24 Gradient boosting; boosting to uniformity.

12:30 lunch

Advanced track: seminars #3

Convener: Tatiana Likhomanenko (National Research Centre Kurchatov Institute (RU))

25 Meta Algorithms Applications in HEP.

seminar3_part1.pdf

15:15 coffee break

26 Hypotheses testing.

seminar3_part2.pdf

Introductory track: seminars #3

Convener: Nikita Kazeev (Yandex School of Data Analysis (RU))

27 Ensemble algorithms: random forest, gradient boosting

3.1 Ensemble methods - seminar.ipynb

COMET_gradient_boosting_gauss_opt.csv

15:15 coffee break

28 Hyperparam Optimization

3.2 Hyperparameters optimzation - seminar.ipynb

17:10 break

29 MVA vs cut-and-count techniques in the SM model processes searches and measurements

Speaker: Lesya Shchutska (University of Florida (US))

SUSY_mva.pdf

Sunday 30 August

08:15 breakfast

Advanced track: lectures #4

Convener: Dr Victor Kitov (MSU)

30 Kernelized algorithms: SVM, regression, K-NN, PCA and their properties.

Kernel trick. Deep learning.pdf

10:20 coffee break

31 Deep learning. Different kinds of learning.

Kernel trick. Deep learning.pdf

Learning Deep Architectures for AI - Bengio.pdf

Introductory track: lectures #4

Convener: Alexey Rogozhnikov

32 GBDT tuning, reweighting, testing hypotheses, gaussian processes for optimization

lecture4.pdf

10:20 coffee break

33 Unsupervised learning, Latent Dirichlet Allocation, PCA. Fast predictions and pruning

12:00 lunch

Advanced track: seminars #4

Convener: Tatiana Likhomanenko (National Research Centre Kurchatov Institute (RU))

34 Summary: HEP ML analysis sketch, from data to discovery.

seminar4_summary.pdf

14:30 coffee break

35 My own first NN using theano.

Introductory track: seminars #4

Convener: Nikita Kazeev (Yandex School of Data Analysis (RU))

36 PCA, RBM and alike on MNIST

4.1 Generative models - seminar.ipynb

14:30 coffee break

37 Clustering

4.2. Clustering - seminar.ipynb

16:15 break

38 Multivariate techniques in the Higgs search

Speaker: Josh Bendavid (CERN)

mlhep-Aug30-2015.pdf

18:00 transfer to the boat

19:00 channel cruise + closing dinner + awards

22:00 transfer to the hotel