Spectral Persistent Homology: Persistence Signals

In this paper, we present a novel family of descriptors for persistence diagrams, reconceptualizing them as signals in $\mathbb R^2_+$. This marks a significant advancement in Topological Data Analysis. Our methodology transforms persistence diagrams into a finite-dimensional vector space through functionals of the discrete measures induced by these diagrams. While our focus is primarily on frequency-based transformations, we do not restrict our approach exclusively to this types of techniques. We term this family of transformations as $Persistence$ $Signals$ and prove stability for some members of this family against the 1-$Kantorovitch$-$Rubinstein$ metric, ensuring its responsiveness to subtle data variations. Extensive comparative analysis reveals that our descriptor performs competitively with the current state-of-art from the topological data analysis literature, and often surpasses, the existing methods. This research not only introduces a groundbreaking perspective for data scientists but also establishes a foundation for future innovations in applying persistence diagrams in data analysis and machine learning.