Writing a Thesis at Our Group

Please do not hesitate to contact any member of our group if you are interested at writing a thesis at our group. 

Requirement:

• You should have successfully taken one of our classes (lecture/seminar) when you want to write a thesis with us.

Completed Theses (at University of Düsseldorf)

Bachelor Theses (2015-2022):

• Active pre-training with phasic policy gradient
• Classification models for argument recommender systems
• Phase retrieval with attention
• Deep learning of financial market dynamics
• Multi-stage progressive image dehazing
• Contrastive self-supervised pretraining of vison transformers
• Erweiterung und Evaluation der Neural Power Unit
• Pay attention to what you calculate: transformer-based approach on recognition of handwritten mathematical expressions
• Ein nicht-operatorbasierter Ansatz für das Conditional Kernel Mean Embedding
• Using unrolled networks for reference based Fourier phase retrieval
• Echtzeiterkennung mit YOLO
• Dynamically modifying ML programs for automated machine learning
• Learning by self-play in turn-based environments
• Natural gradient boosting for classification
• Deep cascading Fourier phase retrieval
• Post-hoc model interpretability vs. intrinsic model interpretability
• Implementing and benchmarking various classes of normalizing flows
• Graph-based semi-supervised leraning with GPs
• Graph-based semi-supervised learning: the distribution of labels
• Learning to write to learn to read
• MRI contrast mapping using machine learning
• Lottery ticket hypothesis - seeking capable subnetworks
• Integrating the Game CATAN into the RL Framework OpenSpiel
• Machine Learning auf Sätzen und ihren semantischen Frames
• Konsistente Kernel Erwartungswert-Schätzung für Funktionen von Zufallsvariablen
• Implementing Continuous High-Resolution Image Reconstruction using Patch Priors
• Adversarial attacks on capsule networks
• Implementing survey propagation
• Optimization of submodular functions
• Nicht-lineare ICA mit neuronalen Netzen
• Chatbots with deep learning
• Implementing AlphaZero for small board games
• Playing Go with Recurrent Neural Networks
• Actor-critic reinforcement learning with experience replay
• Proximal policy optimization (PPO)
• Reimplementing and extending Tesauro's TD-Gammon
• A ML approach to detect and classify spores in microscopy images
• Actor-Critic Reinforcement Learning
• Analyzing Brain Images with Deep Learning
• Deep Q-Learning in TensorFlow
• Grade prediction with machine learning
• Implementation of variational autoencoder
• Classification of data from the ATLAS experiments 
• Collaborative filtering 
• Causal relations for two random variable
• Representing distributions as mixture of Gaussians
• Finding stars in traces      

Master Theses (2015-2022):

• Why don't we have robust classifiers?
• Could Brothers Grimm Create a Dictionary with BERT?
• Topic-aware approaches to Natural Language Processing
• Latent optimization for deep generative phase retrieval
• Cyclophobic reinforcement learning
• Erweiterung von fortschrittlichen Reinforcement Learning Algorithmen
• Classical and integer linear programming approaches to learning causal structure
• Towards better understanding stochastic gradient descent for deep learning
• World models for reinforcement learning
• Measuring the similarity of arguments with BERT
• Scaling deep reinforcement learning
• Deep learning methods for phase retrieval
• Alpha matting revisited
• Capsules for generative adversarial networks
• Bayesian methods for deep learning
• MCMC for Bayesian computation in causal inference
• Multiframe blind deconvolution with lots of noise
• Probabilistic programming in Julia