This shows you the differences between two versions of the page.
Next revision | Previous revision | ||
2019:groups:tools:mlregression [2019/06/22 18:04] roberto.ruiz_de_austri created |
2019:groups:tools:mlregression [2019/06/25 10:38] sascha.caron |
||
---|---|---|---|
Line 1: | Line 1: | ||
== Machine Learning: Regression and classification == | == Machine Learning: Regression and classification == | ||
- | Aims: | ||
- | * Library of regression and classification networks | ||
- | * Keras/Tensorflow and Scikit-learn to start | ||
- | * Library of Training data, maybe connected to hep-data (database) | ||
- | Train regression on the following: | + | See protected pages |
- | * electroweak cross-section in pMSSM19 (DeepXs, ...) | + | |
- | * cross sections Wprime, Zprime (how many dimensions?) | + | |
- | * Doublet Higgs model (6 dim) | + | |
- | + | ||
- | Train likelihoods: | + | |
- | * global fits Gambit Xenoda data (train on likehoods on different observables, e.g. only colliderbit) | + | |
- | * Higgs model with many nuisance parameters (100?) | + | |
- | + | ||
- | Train on observables: | + | |
- | * example relic density | + | |
- | + | ||
- | Train on reconstruction efficiencies: | + | |
- | * LLSps | + | |
- | + | ||
- | (want to share code for the networks...) | + | |
- | + | ||
- | + | ||
- | Output: | + | |
- | - git repositories for networks | + | |
- | - Xenodo data | + | |
- | - talk with hepdata | + | |
- | - paper | + | |
- | - prototype code for the network library | + | |
- | + | ||
- | + | ||
- | Other info: | + | |
- | - How to sample ? Depends on goal, proposed active learning etc. | + | |
- | + | ||
- | + | ||