Chair of Artificial Intelligence and Machine Learning
print
Breadcrumb Navigation
Content

Efficient Optimization of Hierarchical Multi-Label Classifier Ensembles (Ba/Ma)

Topic for a bachelor/master's thesis

Short Description:

In [1] and [2], the authors have demonstrated that a hierarchical decomposition of a multi-label classification problem is beneficial in terms of both efficiency and predictive accuracy. The decomposition in both works is mainly based on relatively simple heuristics. Furthermore, in [3], an approach for optimizing hierarchical structures was proposed. The aim of this thesis is to extend the approach of [3] to cope with the relevant structures in [1] and [2]. Other approaches are also conceivable. Furthermore, the thesis should be centered around the following research questions:

  • Is optimizing the hierarchical structure for decomposing the original multi-label classification problem beneficial?
  • Is it possible to optimize the hierarchical structures in feasible amount of time?
  • Is the choice of base learners an important additional variable for optimization?

Prerequisites

Good background in machine learning, especially supervised learning (e.g., classification, regression) and experimental evaluation, algorithm design, programming skills.

Contact

Marcel Wever

References

  • [1] Tsoumakas, Grigorios, Ioannis Katakis, and Ioannis Vlahavas. "Effective and efficient multilabel classification in domains with large number of labels." Proc. ECML/PKDD 2008 Workshop on Mining Multidimensional Data (MMD’08). Vol. 21. 2008.
  • [2] Jasinska-Kobus, Kalina, et al. "Probabilistic label trees for extreme multi-label classification." arXiv preprint arXiv:2009.11218 (2020).
  • [3] Wever, Marcel, Felix Mohr, and Eyke Hüllermeier. "Ensembles of evolved nested dichotomies for classification." Proceedings of the Genetic and Evolutionary Computation Conference. 2018.