Deep Learning

Contact

Name

Philipp Soldin

Double Chooz

Phone

work
+49 241 80 27323

Email

E-Mail
 

Deep Learning

You can find an institute-wide list of all thesis topics that require deep learning methods and tools on this page. Links to the information of the corresponding supervising research groups are given. A list of "general" topics are given additionally. These are topics that focus on the development of novel deep learning techniques rather than specific applications in research groups. These topics can be worked on either within the framework of a research group or across multiple groups. All topics (and methods) are to be taken as suggestions and can be adapted if necessary. Details on the theses can be discussed with the respective supervisor. For a first impression, you will find a list of deep learning technique keywords attached to each topic. Please note that not all topics may be recommendable for deep learning beginners.

 

Abbreviations

NN: Neural Network

DNN: Deep Neural Network

FCN: Fully Connected Network

CNN: Convolutional Neural Network

GNN: Graph Neural Network

GCN: Graph Convolutional Network

BNN: Bayesian Neural Network

1D Conv.: 1 Dimensional Convolution

GAN: Generative Adverserial Network

 

Bachelor

EnEx/TRIPLE

Acoustic Signal Analysis with Machine Learning Methods

Double Chooz

Parallelization of likelihood calculations with GPUs (GPU programming)

IceAct

Analysis of air-shower images with machine learning methods (Image recognition)

IceCube

Testing improvements of new mashine learning methods and observables for measurements of astrophysical neutrinos (FCNs)

Visualization and Analysis of Feature Importance for Deep Learning Reconstruction Methods (CNN/Feature Analysis)

Testing the robustness of deep learning reconstruction methods against adversarial attacks (Adverserial Attacks)

JUNO

Particle Identification using Deep Learning Methods (GCNs/AutoEncoder/Classic/Recurrent)

SiFi-CC

Improvement of event identification for the SiFi-CC with neural networks (Image recognition/CNNs/FCNs/BNN)

General

  • Discrimination with AutoEncoder (e.g. Positron - Elektron Discrimination) (AutoEncoder)
  • Test of a noval neural network algorithm: GNN based AutoEncoder (e.g. DoubleChooz) (AutoEncoder/GNNs)
 

Master

Einstein Telescope

In data analysis we want to participate in the alert system for multi-messenger triggers. We are seeking master candidates to develop a framework for the identification of inspiralling binary systems from the ET data stream. We want to focus on AI methods, particularly deep learning. Identification will be the pioneering step. We want to extent later to sky localization and parameter reconstruction for the selection of the most interesting events for multi-messenger observation. (Recurrent/Transformer/Attention/AutoEncoder/1D Conv.)

IceAct

Event reconstruction with Machine Learning (Random Forrest, Graph NN)

IceCube

Understanding hidden and correlated uncertainties and Deep Learning Methods in IceCube in the context of adversarial attacks (Adverserial Attacks)

Application of Recurrent Neural Networks for Analysing time series data in IceCube and application to magnetic monopole searches (Recurrent NNs)

Optimizing the DNN Data Selection for the measurement of astrophysical neutrino fluxes (DNN)

JUNO

Particle Identification using Deep Learning Methods (GCNs/AutoEncoder/Classic/Recurrent)

SiFi-CC

Improvement of event identification for the SiFi-CC with neural networks (Image recognition/CNNs/FCNs/BNN)

General

  • Liquid Neural Networks (Liquid NN)
  • Discrimination with AutoEncoder (e.g. Positron - Elektron Discrimination) (AutoEncoder)
  • Feature importance visualization (Feature Analysis)
  • Generative Adverserial Networks (z.B. DoubleChooz/IceCube/Einstein) (GANs (GCNs))