Data for structural semantics using topological data analysis

• This dataset is designed to explore the advantages of utilizing topological summaries of raw data (an abstraction) rather than the raw data itself in machine learning applications.
• This dataset consists of $600$ raw data examples derived from the MNIST handwritten digits dataset: Hand_Digits_Data.
• The labels correspond to the $1$-dimensional holes that appeared in the raw data:Hand_Digits_Data-->Label_Info-->Lablels_v3.

Reference

• From Raw Data to Structural Semantics: Trade-offs among Distortion, Rate, and Inference Accuracy (https://arxiv.org/abs/2412.19825).