Project in Computer Science 1 - Machine Learning Algorithms and Applications

General information

Format

In this course, you will experience the role of a typical machine learning researcher in a research group. You will:

There are three types of projects:


Working on the project consists of the following steps:

1. Problem statement

Choose a project topic or come up with your own, original idea. Then, write a short problem statement, including your research question, motivation, related work, methodology, and expected results.

2. Literature review

Before diving into math and implementation, you will do a literature review of related articles. In a presentation in front of a larger group, you will present an overview over the area of your project and show how your project relates to existing approaches.

3. Working phase

You will then independently work on your project. However, you will have regular discussion meetings with your supervisor and regular progress presentations in front of a larger group of students and supervisors.

4. Final submission

In a final report you will present your approach, the results of your work, and your literature review.

Projects (Tentative)

We are happy to supervise machine learning related projects that are connected to our research interests. Examples are:

Deep Learning in Cataract Surgery (click to expand)
Can GNNs replace transformers or CNNs? (click to expand)
Contrastive learning for GNNs (click to expand)
Convexity in real-world graphs (click to expand)
Empirical investigation of generalization bounds for GNNs (click to expand)
Leakage from Gradients in GNNs (click to expand)
Robustness vs. privacy in GNNs (click to expand)
Zero-One Laws of Graph Neural Networks (click to expand)
Finding Graphs where Message Passing Neural Networks Fail (click to expand)
Privacy attacks on GNNs (click to expand)
Kernel Methods for Large Scale Graph Learning Problems (click to expand)
Learning to logically explain GNNs (click to expand)
Debiasing graph-based models (GNNs) (click to expand)
Removing unwanted bias/information in deep neural networks using Information Theory (click to expand)
Debiasing deep models using Siamese Neural Networks (click to expand)
Disentangled Representations (click to expand)
Reproducibility of ML papers (click to expand)
Robust Machine Learning (click to expand)
Scalable Interactive Analysis (click to expand)
Shortest Path Distance Estimation (click to expand)
Theory of Reinforcement Learning (click to expand)
Safety-Critical Applications of Reinforcement Learning (click to expand)

Your own idea!

Describe the scientific merit and novelty of your idea. It is very important to narrow down the rough topic to a tentative research question and approach of interest to us. The research question should not have been answered previously and the answer needs to be verifiable. To answer the question, typically one has to:

If you choose your own topic, please briefly describe your project following this structure (check our suggested topics to get an idea):

General resources (freely available books and lecture notes)