Concept-based explanations have emerged as an approach to make models explainable in a human-understandable way. In this article, we investigate extracted concepts for a music emotion recognition model. In a listening experiment, we explore properties of found concepts and show how to present them to users. The article is based on our paper and introduces only parts of our research. More information is provided in the paper.
© Verena Szojak & Verena Praher
Figure 10: Word clouds summarising the descriptions that participants gave for Layer4 CAV9: no clear semantic focus can be identified.
No article content yet.
Verena Praher
Verena Praher is a Data Scientist with a background in Machine Learning, Explainable Artificial Intelligence and Sound Processing. During her PhD studies at the Institute of Computational Perception at JKU Linz, she focused on making explanations for deep models in Music Information Retrieval understandable for humans.
Verena Szojak
Verena Szojak is studying Artificial Intelligence and is currently completing her master's degree at JKU Linz. In her bachelor's thesis, she investigated approaches to overcome the limitations of concept-based explanations for an acoustic scene classifier.
