[1Win4] Poster session 1

[1Win4-02] Investigating the Manifold Hypothesis in Deep Learning through Dimensionality Reduction Methods

〇Chika Obata¹, Kenya Jinno¹ (1.Tokyo City University)

Keywords:manifold hypothesis, auto encoder, principal component analysis

This study examines the effectiveness of dimensionality reduction methods in deep learning based on the manifold hypothesis. To analyze the latent structure of high-dimensional data, Principal Component Analysis (PCA) and Autoencoders were applied, and criteria for optimal dimensionality selection were evaluated based on classification performance. Experiments using the MNIST dataset suggest that the features of MNIST can be adequately represented in a 17-dimensional space, enabling efficient data representation through appropriate dimensionality reduction. Future research will focus on validating these findings with more complex datasets and comparing various dimensionality estimation techniques, such as Manifold Fit-based Tangent Manifold Analysis (MFTMA), to establish a more general criterion for dimensionality selection.