%new physics at the lhc %introduce toptagging on this slide Classify events into those that origin from a top quark, and those by other qcd particles to do this, use either calorimeter like images or 4-vectors classical approach: first build a theory (for example super symmetry) make predictions test them not very effective in the last time so try using unsupervised algorithms to find 'weird' stuff these algorithm are tested quite well using top tagging since the top quark was only discovered 1995, so before this, tops actually were 'weird' the top quark has a quite low cross section (about #1# top event for each #10# million collisions) %slide to show the history of toptagging classically you use smart physics to differentiate them (arXiv:1806.01263) but then there were deep learning approaches (arXiv:1704.02124) which do this a bit better today even better using a fancy graph neuronal network (ParticleNet,arXiv:1902.08570) Supervised Training given both the anomaly and the background events Much easier to do only able to find one specific anomaly Unsupervised Training only given background events Able to find any anomaly Used by QCDorWhat (arxiv 1808.08979) for unsupervised toptagging