Dimension Reduction using PCA and GPLVM for Simulated Cylinder and Real Handwritten Digits

This project explores dimensionality reduction techniques through a comparative analysis of Principal Component Analysis (PCA) and Gaussian Process Latent Variable Models (GPLVM).
The methods were applied both to a simulated cylindrical dataset and to the MNIST handwritten digits dataset to evaluate and compare their effectiveness in capturing low-dimensional structures within high-dimensional data.

The work represents my bachelor’s thesis project, focused on assessing the strengths and limitations of linear (PCA) and nonlinear (GPLVM) approaches for feature extraction and visualization.

Download the Thesis PDF