Another Topological "Reading" Lesson: Classification of MNIST using Bottleneck-based Statistical Features

Abstract: Inspired by the work of Garin and Tauzin on the classification of grayscale images using features from a wide array of topological summaries, we perform classification of the MNIST data set using only features derived from statistics on bottleneck distances. While there exist several critiques on the bottleneck metric, we show that it can be used to produce features on which machine learning algorithms, in this case a random forest, can be trained to produce respectable accuracy.

computational geometryalgebraic topologycombinatoricsgeometric topologyprobability

Audience: advanced learners

Asia Pacific Seminar on Applied Topology and Geometry