Thesis

Big data, Machine Learning and object classification in high energy hadronic collisions

• Call:

  IDPASC Portugal - PHD Programme 2017

• Academic Year:

  2017 / 2018

• Domains:

  Theoretical Particle Physics | Experimental Particle Physics

• Supervisor:

  José Guilherme Milhano

• Co-Supervisor:

  Nuno Castro

• Institution:

  Instituto Superior Técnico

• Host Institution:

  LIP, IST and UMinho

• Abstract:

  Machine Learning (ML) has become pervasive in today's world. Web search engines, image and voice recognition, targeted advertising and personalized music/film recommendations, credit card fraud detection and email spam filters all rely heavily neural network architectures. Efficient training of an artificial neural network typically requires very large amounts of data: input/output pairs from which the network will learn to predict the output for new inputs. The extremely large amounts of data generated by particle colliders make the use of ML both a necessity and a potentially very fruitful path to follow. ML techniques are used extensively in many areas of high energy particle physics with application ranging from low level tasks, such as identification of physical objects in collider data (top quarks, W and Z bosons, tau, ...) to high-level physics analyses discriminating between specific and rare signals and known backgrounds. More recently, ML has proved to be a powerful physics discovery tool allowing to identify important properties of physical objects (e.g. QCD jets) from 'detector-level' information that had escaped the imagination of theorists. This thesis will have a dual focus: (i) the application of ML to the efficient identification of physical objects in proton-proton, proton-nucleus and nucleus-nucleus collisions at LHC and future collider energies; (ii) development of systematic methods to learn Physics from the ML, that is to identify what is learnt by the machine and match to either existing analytical calculations or carry on those calculations. All work will be carried out using both Monte-Carlo generated samples and Open (publicly available) LHC data. The selected candidate will develop both strong and highly transferable computational skills and solid competence in Quantum Chromodynamics. The thesis will be co-supervised by an experimentalist (Nuno Castro) and a phenomenologist/theorist (Guilherme Milhano) and the selected candidate will divide her/his time between Braga and Lisbon.