indico of the meeting:
<div>
    https://indico.cern.ch/conferenceDisplay.py?confId=231636</div>
<div>
    &nbsp;</div>
<div>
    CMS talk (M. Malberti):</div>
<div>
    ------------------</div>
<div>
    &nbsp;</div>
<div>
    - Content:</div>
<div>
    &nbsp; &nbsp; a set of plots which show the effect of the MVA selection on dijet</div>
<div>
    &nbsp; &nbsp; e.g. basic jet quantities + ptjj, deltaphi(jj,gg), pt(jjgg)</div>
<div>
    &nbsp;</div>
<div>
    - Questions and comments:</div>
<div>
    &nbsp; &nbsp;* what powheg? R: normal powheg, no minlo</div>
<div>
    &nbsp; &nbsp;* please redo the plot for ggH. R: done and added into the new</div>
<div>
    &nbsp; &nbsp; &nbsp;version of the talk</div>
<div>
    &nbsp; &nbsp;* how do you prevent to not have deltaphi(gg,jj) values close to pi?</div>
<div>
    &nbsp; &nbsp; &nbsp;In atlas we impose that MVA do not use deltaphi(gg,jj) variables</div>
<div>
    &nbsp; &nbsp; &nbsp;above a given threshold. That make efficiency in that region ~flat.</div>
<div>
    &nbsp; &nbsp; &nbsp;R: we don&#39;t. We will check the efficiency behavior.</div>
<div>
    &nbsp;</div>
<div>
    &nbsp;</div>
<div>
    1) ATLAS talk (D. Gillberg):</div>
<div>
    ---------------------</div>
<div>
    &nbsp;</div>
<div>
    - Content</div>
<div>
    &nbsp; &nbsp; a set of plots which show the effect of the MVA selection on dijet</div>
<div>
    &nbsp; &nbsp; e.g. basic jet quantities + ptjj, deltaphi(jj,gg), pt(jjgg)</div>
<div>
    &nbsp;</div>
<div>
    - Question and comments:</div>
<div>
    &nbsp; &nbsp;* why do you evaluate systematics at 95% efficiency point? R:</div>
<div>
    &nbsp; &nbsp; &nbsp;arbitrary choice, only to get a cross-check what to expect from method</div>
<div>
    &nbsp; &nbsp; &nbsp;shown in later in the other talk</div>
<div>
    &nbsp; &nbsp;* would be useful to have efficiency plots for both CMS and ATLAS</div>
<div>
    &nbsp; &nbsp;* imposing that MVA does not use deltaphi(gg,jj) above a given</div>
<div>
    &nbsp; &nbsp; &nbsp;threshold reduces separation power. Maybe a trade-off&nbsp;</div>
<div>
    &nbsp; &nbsp; &nbsp;between loss in separation and theory uncertainty would&nbsp;</div>
<div>
    &nbsp; &nbsp; &nbsp;be beneficial. R: not for now but interesting for future studies</div>
<div>
    &nbsp;</div>
<div>
    &nbsp;</div>
<div>
    2) ATLAS talk (F. U. Bernlochner):</div>
<div>
    --------------------------</div>
<div>
    &nbsp;</div>
<div>
    - Content</div>
<div>
    &nbsp; &nbsp; development of a new technique to evaluate systematics due to</div>
<div>
    &nbsp; &nbsp; radiation of extra jets in ggH for dijet channel when using MVAs.&nbsp;</div>
<div>
    &nbsp; &nbsp; Currently ST-like uncertainties used as input (technique itself can</div>
<div>
    &nbsp; &nbsp; be used with any input uncertainties)</div>
<div>
    &nbsp;</div>
<div>
    - Question and comments:</div>
<div>
    &nbsp; &nbsp; * how do you deal with MCFM for regions with negative cross</div>
<div>
    &nbsp; &nbsp; &nbsp; sections? R: Currently, no shape information is used below a certain</div>
<div>
    &nbsp; &nbsp; &nbsp; cut but only normalization. (In principle we can extend the method in</div>
<div>
    &nbsp; &nbsp; &nbsp; that region but once MCFM-only uncertainties are &gt; 100%</div>
<div>
    &nbsp; &nbsp; &nbsp; it becomes difficult to interpret the result.)</div>
<div>
    &nbsp; &nbsp; * are there more checks that the method works, even if it uses</div>
<div>
    &nbsp; &nbsp; &nbsp; categorization based on a single variable? page 11 looks like a</div>
<div>
    &nbsp; &nbsp; &nbsp; trivial closure test. R: it is the test at page 9</div>
<div>
    &nbsp; &nbsp; * why is the uncertainty in page 9 different between pure MCFM</div>
<div>
    &nbsp; &nbsp; &nbsp; and powheg reweighted? R: spectra in the two MCs are different, we</div>
<div>
    &nbsp; &nbsp; &nbsp; were actually happy to see this agreement</div>
<div>
    &nbsp; &nbsp; * why not doing it on the MVA directly, i.e. by applying MVA to MCFM&nbsp;</div>
<div>
    &nbsp; &nbsp; &nbsp; ntuple output more than via deltaphi method?</div>
<div>
    &nbsp; &nbsp; &nbsp; R: technically more involved, maybe using MVA directly on MCFM would</div>
<div>
    &nbsp; &nbsp; &nbsp; be important to understand if it makes a difference if the uncertainty</div>
<div>
    &nbsp; &nbsp; &nbsp; is 35% vs. 40% vs. 45% but given the present statistical</div>
<div>
    &nbsp; &nbsp; &nbsp; uncertainty of this channel would not make such a difference</div>
<div>
    &nbsp; &nbsp; * is it possible to get weights from ATLAS and apply them to CMS</div>
<div>
    &nbsp; &nbsp; &nbsp; montecarlo? R: good idea to have consistency.</div>
<div>
    &nbsp; &nbsp; &nbsp; Weights could be ready by the beginning of next week.</div>
<div>
    &nbsp;</div>
<div>
    ACTION ITEMS:</div>
<div>
    ------------</div>
<div>
    &nbsp; &nbsp;* Experiments should discuss on how to proceed to have a common</div>
<div>
    &nbsp; &nbsp; &nbsp;approach and in particular have numbers which can be directly compared.</div>
<div>
    &nbsp; &nbsp;* The technique developed by Florian and Dag (ATLAS) and using inputs<br>
    &nbsp; &nbsp; &nbsp;from F. Tackmann et al. paper can be used by CMS physicists<br>
    &nbsp; &nbsp;* There is the Issue that ATLAS is imposing flat efficiency above a</div>
<div>
    &nbsp; &nbsp; &nbsp;given threshold in deltaphi and the deltaphi bins are &nbsp;based on</div>
<div>
    &nbsp; &nbsp; &nbsp;that. CMS can check the efficiency above that value.</div>
<div>
    &nbsp; &nbsp;* Maybe the starting point is to apply it to the Hgg channel with deltaphi</div>
<div>
    &nbsp; &nbsp; &nbsp;as baseline variable</div>
<div>
    &nbsp; &nbsp;* Once enough experience other channels can follow (using the</div>
<div>
    &nbsp; &nbsp; &nbsp;appropriate baseline variable for them, e.g. pTHjj)</div>
<div>
    &nbsp; * Code from ATLAS can be ready by beginning of next week and passed&nbsp;</div>
<div>
    &nbsp; &nbsp; &nbsp;to the LHC working group which will make it public (with</div>
<div>
    &nbsp; &nbsp; &nbsp;documentation to follow)<br>
    &nbsp;</div>