Topological Data Analysis of MNIST Dataset

A novel approach to machine learning tasks on grayscale images is explored and implemented by integrating Topological Data Analysis (TDA). The method applies persistent homology to generate a wide array of topological features from a point cloud derived from the image, using grayscale filtration and various filtrations defined on the binarized image. The effectiveness of this pipeline is demonstrated as a significant dimensionality reduction tool on the MNIST digits dataset, achieving similar accuracy in digit image classification.

Project Poster