Machine learning is used to reconstruct individual particles in the detector and also to separate signal processes from background processes.
A general example... process A and process B can both have electrons in their final states (with other objects...). ML is used to separate A and B based on the kinematic properties of electrons (in combination with those other objects). Also, higher upstream, ML was used just to know that we had an electron to begin with!
A lot of BDTs are used, with deep learning under very active investigation. For example, when looking for the Higgs decaying to two bottom quarks, a slew of ML algorithms are used to identify so-called "b-jets" (jets which are identified as originating from a b-quark). In ATLAS we have low level taggers using deep neural networks (using Keras) in combination with higher level taggers using BDTs. Another example is the recent ttH observation, where XGBoost was used [1].
A general example... process A and process B can both have electrons in their final states (with other objects...). ML is used to separate A and B based on the kinematic properties of electrons (in combination with those other objects). Also, higher upstream, ML was used just to know that we had an electron to begin with!
A lot of BDTs are used, with deep learning under very active investigation. For example, when looking for the Higgs decaying to two bottom quarks, a slew of ML algorithms are used to identify so-called "b-jets" (jets which are identified as originating from a b-quark). In ATLAS we have low level taggers using deep neural networks (using Keras) in combination with higher level taggers using BDTs. Another example is the recent ttH observation, where XGBoost was used [1].
[1] https://atlas.cern/updates/physics-briefing/observation-tth-...