S. Caron, J.S. Kim, K. Rolbiecki, R. Ruiz de Austri and B. Stienen,
The BSM-AI project: SUSY-AI - Generalizing LHC limits on Supersymmetry with Machine Learning
[arXiv:1605.02797]

About SUSY-AI Online

Physics and machine learning

This page gives a short summary of the science behind (this online interface of) SUSY-AI. It aims at quickly creating a feeling for the physics of the program and the output of the program. This approach may leave some open questions. For a full description of SUSY-AI and the physics described by it, the reader is refered to our research paper:

S. Caron, J.S. Kim, K. Rolbiecki, R. Ruiz de Austri and B. Stienen,
The BSM-AI project: SUSY-AI - Generalizing LHC limits on Supersymmetry with Machine Learning
[arXiv:1605.02797]

SUSY-AI Online contains a machine learning algorithm. By learning the patterns of model exclusion in the 19-dimensional pMSSM parameter space, this algorithm is able to predict the exclusion of any model point in this model space as if it were calculated by a full exclusion analysis by the ATLAS collaboration. However, since training is inherently done on a finite set of model points and the basis of machine learning is formed by statistical methods, the output of the algorithm is subject to statistical fluctionations. In our paper (see link above) we have however shown that the trained algorithm is able to make predictions with 93.2% accuracy. This accuracy can be enhanced by requiring a minimum confidence level: model points with a minimum confidence level of 0.95 for example have an accuracy on their prediction of over 99%.

The input for SUSY-AI Online is a set of 19 input parameters related to the soft breaking of the supersymmetric spectrum. The variables in this set are labeled as they would be in .slha files. For completeness a list of these variables (each a short explanation and their default location in .slha files) can be found in the table below.

Variable About .slha BLOCK .slha SWITCH
M1 Bino mass parameter MSOFT 1
M2 Wino mass parameter MSOFT 2
M3 Gluino mass parameter MSOFT 3
mL1 First and second generation left-handed slepton breaking mass MSOFT 31
mL3 Third generation left-handed slepton breaking mass MSOFT 33
mE1 First and second generation right-handed slepton breaking mass MSOFT 34
mE3 Third generation right-handed slepton breaking mass MSOFT 36
mQ1 First and second generation left-handed squark breaking mass MSOFT 41
mQ3 Third generation left-handed squark breaking mass MSOFT 43
mU1 First and second generation right-handed up-type squark breaking mass MSOFT 44
mU3 Third generation right-handed up-type squark breaking mass MSOFT 46
mD1 First and second generation right-handed down-type squark breaking mass MSOFT 47
mD3 Third generation right-handed down-type squark breaking mass MSOFT 1
At Trilinear stop Yukawa coupling AU or TU 3,3
Ab Trilinear sbottom Yukawa coupling AD or TD 3,3
Atau Trilinear stau Yukawa coupling AE or TE 3,3
mu Higgsino mass parameter HMIX 1
MA^2 Pseudoscalar Higgs mass (squared) HMIX 4
tan(beta) Ratio of vacuum expectation values of H^0_u and H^0_d HMIX 2

Talks

The SUSY-AI (Online) collaboration has given various talks on the program and (online) interface. A list of all talks and slides used in them are (if applicable) to be found here.


Research

SUSY-AI is a webinterface to an instance of SUSY-AI: a Python software package mimicking ATLAS exclusion limits. These exclusion limits are generated with machine learning: having learned the exclusion of over 300,000 model points in the pMSSM, it is able to predict the exclusion of unseen model points with a minimum accuracy of 93.2%. Papers describing the data that was used in this training can be found here. For a full description of the use of machine learning in the generalisation of LHC exclusion limits, see the original paper:


Webapplication

SUSY-AI Online is a webinterface to an instance of SUSY-AI: a Python package that contains all functionalities of SUSY-AI Online. The advantage of using the package locally however is that it can then also perform batch predictions, allowing for exclusion determination of 10,000 per second.

The software running on the server is a version of SUSY-AI that is optimized for running in webapplication-like environments. As a consequence, the SUSY-AI Online is considerably slower than the original SUSY-AI package. If the user requires faster predictions or predictions on large batches of spectrum files/coordinates, downloading SUSY-AI and running it locally is recommended.
More information on SUSY-AI (i.e. on how to install and use it) can be found on the project website: http://susyai.hepforge.org/


Changelog

All notable changes to this project will be documented here. The format is based on Keep a Changelog (http://keepachangelog.com/). Please note that the versioning of SUSY-AI Online is independent of the versioning of SUSY-AI.

##[2.3.0] - Thu 2017-07-13 ### Added - Game: Challenge the Machine! ### Changed - Version indication in header now indicates correct SUSY-AI version (instead of SUSY-AI online version) ## [2.2.1] - Tue 2017-04-18 ### Changed - PySLHA (used to read .slha files) has been updated to version 3.2.0 ## [2.2.0] - Wed 2017-04-12 ### Added - Website now remembers which predict type was selected on refresh of webpage ### Changed - SUSY-AI version updated to 1.1.5 (most recent one) This version fixes a mapping bug in which negative masses would map on the low positive range (they map now to the high positive range) - Sliders now have a darker gray background color when selected value is out of parameter range ## [2.1.1] - Mon 2017-01-09 ### Fixed - Bug that messed with the outlier labeling ## [2.1.0] - Mon 2016-10-17 ### Changed - Behind the screen behaviour has been optimized for situations in which only subset of classifiers is online ### Fixed - Critical bug preventing starting of the service ## [2.0.5] - Tue 2016-09-27 ### Added - Explanation of the input variables (see About page) - Links on the main page to the explanation of the input variables ## [2.0.4] - Thu 2016-09-22 ### Fixed - Slider colour now changes on manual numerical input of parameter value ### Added - Proper warning regarding the possible violation of the LSP / higgs requirement when spectrum was manually entered into the website (via the sliders) - How To was altered to contain the new warning (see addition above) ## [2.0.3] - Thu 2016-09-15 ### Fixed - Responsiveness and website title/favicon fixed for susy-ai.org and susy-ai.com ## [2.0.2] - Mon 2016-09-12 ### Added - susy-ai.com and susy-ai.org now link to this website - Page containing references to papers associated with the used data - Links to above mentioned page, both in sidebar and on ABOUT page ## [2.0.1] - Tue 2016-08-30 ### Changed - Classifier indicators now have correct notation with capital letters and spaces ### Fixed - Some minor textual errors in the How To - Some minor textual errors on the About page ## [2.0.0] - Sat 2016-08-27 ### Added - Multi method analysis: all results are now predicted both for 8TeV and 13TeV analysis - Selection of analysis in colour indication - Information in How To... on how multi method analysis works ### Fixed - Minor bugs in visualisation of results - The 'More information' link errors occasionally return now also opens the How To on the right slide ### Changed - Visuals in How To are now up to date ## [1.2.1] - Wed 2016-08-22 ### Changed - Some minor graphical improvements were implemented ### Fixed - How to interface - Link "More information" about mapping results - The header and SUSY-AI logo link properly to the homepage ## [1.2.0] - Mon 2016-08-20 ### Added - An 'about' page, allowing for a better explanation of the webapplication and software, as for an implementation of a changelog - Apple touch icons

To Do

  • Currently there are no planned updates in the near future.
    Suggestions? Contact us!

Contact

Although both SUSY-AI and SUSY-AI Online have been build with great attention to detail, some errors may still persist in their current releases. If you encounter any of these errors, or if you have questions about the software (or maybe suggestions on how it could be made even better?), don't hesitate to contact us on: b.stienen@hef.ru.nl.
SUSY-AI and SUSY-AI Online were developed by S. Caron, J.S. Kim, K. Rolbiecki, R. Ruiz de Austri and B. Stienen.
If you encounter any problems, don't hesitate to contact us!
SUSY-AI and SUSY-AI Online (c) 2017