Investigation of interpretability approaches for CNN-based classification models and their generalisability across various datasets in tire production.

Darwin Rolando SUCUZHANAY MORA

Diplomová práce

Abstract:

This thesis enhances the interpretability of various Convolutional Neural Network (CNN) models used at different stages of tire manufacturing at Continental. It investigates multiple approaches that can be generalized across various image classification tasks. The methods investigated in this work introduced different levels of interpretability into the CNN models, offering both a global explanation of model behavior through latent space representations and a local explanation of specific predictions using Grad-CAM, Grad-CAM++, and Layer-wise Relevance Propagation (LRP). The explanations generated by the latter methods were evaluated across multiple criteria, including faithfulness, robustness, complexity, and randomization.

Abstract:

This thesis enhances the interpretability of various Convolutional Neural Network (CNN) models used at different stages of tire manufacturing at Continental. It investigates multiple approaches that can be generalized across various image classification tasks. The methods investigated in this work introduced different levels of interpretability into the CNN models, offering both a global explanation of model behavior through latent space representations and a local explanation of specific predictions using Grad-CAM, Grad-CAM++, and Layer-wise Relevance Propagation (LRP). The explanations generated by the latter methods were evaluated across multiple criteria, including faithfulness, robustness, complexity, and randomization.

Keywords

Artificial Intelligence, Explainable Artificial Intelligence, Convolutional Neural Network, Local Explanation, Global Explanation, Faithfulness, Robustness, Complexity, Randomization.

Jazyk práce: angličtina

Datum vytvoření / odevzdání či podání práce: 20. 8. 2025

Obhajoba závěrečné práce

• Vedoucí: Jishnu Seshadri, M.Sc.

JIHOČESKÁ UNIVERZITA V ČESKÝCH BUDĚJOVICÍCH

Přírodovědecká fakulta

Magisterský studijní program / obor:
Artificial Intelligence and Data Science / Artificial Intelligence and Data Science