2020/10/05 by Nikola Milićević, Milicevic, Nikola
Computer Science · Physics and Astronomy · #Topological and Geometric Data Analysis #Advanced Graph Neural Networks #Complex Network Analysis Techniques
paper · pdf · doi:10.48550/arxiv.2010.02020
We conduct a study of real-valued multi-parameter persistence modules as sheaves and cosheaves. Using the recent work on the homological algebra for persistence modules, we define two different convolution operations between derived complexes of persistence modules. We show that one of these operations is canonically isomorphic to the derived tensor product of graded modules. We give formulas for computing convolutions between single-parameter interval decomposable modules. Our convolution operations are analogous to the convolution of derived complexes of constructible sheaves on ℝn introduced by Schapira and Kashiwara. In our setting, ℝn has a non-standard topology. We show our convolution operation satisfies analogous properties to the convolution of constructible sheaves on ℝn with the standard topology. We define a convolution distance for derived complexes of persistence modules and show that it extends the classical interleaving distance. We also prove stability results from the sheaf and cosheaf points of view.