Dimensionality Reduction and Classification in High-Dimensional Data: A Hybrid Approach Using Generalized Matrix LVQ and Deep Learning Techniques

Hadjichristodoulou, Christodoulos (2024) Dimensionality Reduction and Classification in High-Dimensional Data: A Hybrid Approach Using Generalized Matrix LVQ and Deep Learning Techniques. Master's Thesis / Essay, Computing Science.

Preview	Text mCS2024HadjichristodoulouC.pdf Download (4MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (128kB)

Abstract

Recent advancements in neural networks, particularly convolutional neural networks (CNNs), have significantly improved performance in various machine learning tasks. However, the inherent black-box nature of these models often impedes the interpretability of their decision-making processes, which is crucial for gaining deeper insights and ensuring trust in critical applications. This thesis addresses the challenge of explainability in high-dimensional data classification by proposing a hybrid approach that integrates Generalized Matrix Learning Vector Quantization (GMLVQ) with deep learning techniques, specifically autoencoders. The proposed method leverages the strengths of GMLVQ in providing interpretability through prototype-based classification, combined with the dimensionality reduction capabilities of autoencoders. By making use of the decoder part, the approach aims to map the reduced-dimensional space back to the original feature space, thus offering a transparent view of the features influencing the classification outcomes.

Item Type:	Thesis (Master's Thesis / Essay)
Supervisor name:	Biehl, M. and Veen, R.J. and Bunte, K.
Degree programme:	Computing Science
Thesis type:	Master's Thesis / Essay
Language:	English
Date Deposited:	03 Sep 2024 10:31
Last Modified:	03 Sep 2024 10:31
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/34168

Actions (login required)

View Item