Project information

  • Category: Machine Learning
  • Type: Dimensionality Reduction
  • Client/Purpose: This experiment aimed to reproduce the ISOMAP algorithm results from the original paper by Tanenbaum et al.
  • Project date: June, 2020
  • Project URL: github

Dimensionality reduction in high-dimensional visual space

Reproducting the ISOMAP algorithm on a canonical problem in dimensionality reduction for visual perception. The input consists of many images of a person’s face observed under different pose and lighting conditions, in no particular order. These images can be thought of as points in a high-dimensional vector space, with each input dimension corresponding to the brightness of one pixel in the image.

With the implementation of the ISOMAP the resulting 2-dimensional embedding is more interpretable, at least in one dimension. The experiments used both Euclidian and Manhattan distances as similarity measures while choosing a threshold so that each node has at least 100 neighbors.

Images of faces with similar values in the embeddings were chosen to illustrate the clustering.