We offer the following master thesis in our research group:
Introduction into the topic
Random projections serve as an economical means to diminish data dimensionality while often preserving important intrinsic characteristics. Through the utilization of random matrices, this technique efficiently converts high-dimensional data into lower-dimensional representations, enabling expedited computations and simplified analysis. Widely used in machine learning, random projections are increasingly acknowledged as a crucial element within ML frameworks.
- Paper: Random projection (https://cs-people.bu.edu/evimaria/cs565/kdd-rp.pdf)
- sklearn (https://scikit-learn.org/stable/modules/random_projection.html)
- Dissertation about the topic (https://arxiv.org/pdf/1812.09489.pdf)
Task:
The primary objective of this thesis is to assess the significance of random projections in classification tasks. Initially, a short review of existing literature should be conducted to ascertain the current understanding in this field. This involves giving a glimpse in the established knowledge and providing insights into the existing research landscape.
Following the literature review, a simple experiment will be conducted to compare the classification performance using two scenarios: a basic single-layer linear model versus the same model with a random projection layer in front. This experiment aims to elucidate the impact of incorporating random projections on classification accuracy.
Expectation:
- Introduction into the topic Random Projections in the context of Machine Learning
- Developing a simple experiment which tries to investigate the power of random projections in context of classification
- Replicable experiment and reusable Code
- Incorporate visual representations of your findings within your thesis and proper explanation of your findings
Skills Required:
- Strong background in linear algebra and probability theory
- Motivated to conduct independent literature searches
- Basic programming (Pytorch or Tensorflow)
Advantages:
- Engaged in a topic of high interest within our research group
- Likelihood of inclusion in the research paper
Supervisor: Marius Lotz (mailto:marius.lotz@fb2.fra-uas.de)