Bachelorarbeiten bei Prof. Krämer in 2020
Unsupervised learning for LHC physics
Machine learning has played an important role in the analysis of high-energy physics data for decades. Unsupervised deep learning, for example using autoencoders, provides a new method of searching for new physics at the Large Hadron Collider LHC. The key idea of the autoencoder is that it learns to map ordinary particle physics events back to themselves, but fails to reconstruct exotic events from new physics that it has never encountered before. In this Bachelor thesis, you will explore various neural network architectures for unsupervised learning and their application to detect exotic signatures from new physics models at the LHC.
Neural networks for event generation in astroparticle physics
The analysis of cosmic rays is a powerful tool to search for dark matter. The interpretation of cosmic ray data in terms of dark matter signals, however, requires a detailed modelling of the propagation of cosmic rays in our galaxy. The parameter space of models for cosmic ray propagation has a high dimensionality, which is most efficiently dealt with using machine-learning techniques, in particular deep neural networks. In this Bachelor thesis you will develop, train and validate such a neural network in order to substantially reduce the computation time needed for the interpretation of dark matter signals in cosmic rays.
Field theory for neural networks
Many qualitative features of the emerging collective dynamics in neuronal networks, such as correlated activity, stability, response to inputs, chaotic and regular behavior, can be understood in models that are accessible to a treatment in statistical field theory. In particular, functional or path integral formulations of field theory provide compact representations of the physical content of such models, for example in terms of Feynman diagrams, and they allow to describe the collective behavior that emerges from the interaction between phenomena on a multitude of scales. In this Bachelor thesis, you will learn the basics of statistical field theory and derive mean-field approximations to the dynamics of neuronal networks.
Sorting out Higgs decays
The Higgs boson has been discovered almost 8 years ago but the investigation of its detailed properties is still a fascinating endeavour. A specific Higgs-boson measurement is usually performed by looking at a specific set of particles the Higgs boson can decay into, a so-called final state. There are current efforts to define these final states in a mutually exclusive way, so that each Higgs event can be unambiguously classified. In this master thesis, you will use state-of-the-art simulation tools to perform a subset of this classification procedure.