This master thesis aims at developing neural networks for unsupervised learning of features in trademark images. The objective is to leverage the huge numbers of trademarks in our possession, in order to propose a new feature-space embedding, using unsupervised methods such as entoencoders. You will evaluate whether this new approach helps in the performances of trademark retrieval systems.
This subject is interdisciplinary and will require collaborations with two Law Master Thesis students. You will present and explain the different algorithms, and enable them (using, e.g. Explainable AI techniques) to analyse and evaluate the systems for their discipline.
Contact : Thomas Vandamme (thomas.vandamme@ulb.be)
Promoter : Prof. Olivier Debeir
marque d'intérêt: yang.liu@ulb.be