Institutional Repository of Vilnius University
Klasifikavimo uždavinio rezultatų vizualizavimas /
Abstract
dc:descriptionNeural networks, while a powerful tool in data analysis and decision-making, are often referred to as "black boxes" because of their unclear operation. Understanding how a model makes decisions is important to have confidence in the results it produces. The Layer-Wise Relevance Propagation (LRP) approach offers the possibility to explain these model decisions. The aim of this study was to investigate how LRP can be used to visualise neural network decisions in classification tasks. To do so, experiments were carried out with the VGG16 model to solve the brain cancer classification task. Experiments with different epsilon (ε) and gamma (γ) parameters revealed that parameters close to zero provide the best quality of importance maps, allowing a clear view of which brain regions are most important for the model solution. The IoU metric was also used to objectively assess how well the LRP-generated importance maps match the locations of real brain tumours. This study showed that the LRP approach is an effective tool for visualising neural network solutions, but its performance is highly dependent on the weights learned by the classification model.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Raišuotytė, Rosita,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD210647886&prefLang=en_US