Possible Thesis Topics with Oliver Schulte (December 2017)

I am researching machine learning for structured data, SQL, XML, network data, event logs, sports data.

Relational Learning.

I am investigating machine learning for relational and network data. Topics include:
  • Learning Bayesian networks. We use these to model the joint distribution of object attributes and links, and for feature generation.
  • Classification.
  • Anomaly Detection and Exception Mining.
My group has developed novel methods for these problems that work well already. Scaling to big data with millions of datapoint is a particular strength of our methods.
Scaling to hundreds of features (attributes and relationship types) is one of the research topics I would like to work on. This combines systems tools (e.g. Spark, Hadoop) with machine learning.

Application Areas

Relational learning has many exciting application areas. I am especially interested in the following.
  • Statistical Modelling of Sports Data. E.g. player ranking, drafting decisions, match outcome prediction.
  • Detecting Relationships in Computer Vision.
  • Extracting Relationships from Text and Images.
  • Anomaly Detection, Data Cleaning.
  • Business Process Mining (see the BPM challenge ).
I'm also interested in challenge competitions for structured data like the Yelp Dataset Challenge.

Detailed Descriptions

This research proposal outlines a five-year plan for relational machine learning, with motivation, background, references. If you want to see more details about where your thesis might fit in, here is my plan for group collaboration.

Deep Reinforcement Learning for Sports Analytics

In my opinion, reinforcement learning has the concepts and algorithms to solve the problems of sports analytics. I would like to work with a Ph.D. student on extending our current neural network methods to multi-level deep reinforcement learning . By that I mean building models for a class hierarchy of players, e.g. professional player -> professional defensive player -> specific defensive player. Applications include player ranking and perhaps match outcome prediction.