Abstract:The fields of effective resistance and optimal transport on graphs are filled with rich connections to combinatorics, geometry, machine learning, and beyond. In this article we put forth a bold claim: that the two fields should be understood as one and the same, up to a choice of $p$. We make this claim precise by introducing the parameterized family of $p$-Beckmann distances for probability measures on graphs and relate them sharply to certain Wasserstein distances. Then, we break open a suite of results including explicit connections to optimal stopping times and random walks on graphs, graph Sobolev spaces, and a Benamou-Brenier type formula for $2$-Beckmann distance. We further explore empirical implications in the world of unsupervised learning for graph data and propose further study of the usage of these metrics where Wasserstein distance may produce computational bottlenecks.
What problem does this paper attempt to address?
The paper mainly discusses the equivalence between effective resistance and a specific optimal transportation problem on graphs. The authors propose a bold idea that effective resistance and optimal transportation should be regarded as the same thing, just depending on the choice of parameter p. They introduce a parameterized p-Beckmann distance family and compare it with Wasserstein distance, especially on random walks on graphs, graph Laplacian operator, and graph Sobolev space.
In the paper, the authors demonstrate the general properties of p-Beckmann problem, including its duality theory, and provide examples on paths and trees. For the 2-Beckmann problem, the authors conduct in-depth studies from three perspectives: effective resistance of probability measures, negative Sobolev seminorm distance, and linearized optimal transportation distance. Furthermore, they propose a Benamou-Brenier type formula for 2-Beckmann distance and apply it to unsupervised learning tasks, taking graph data classification as an example, showcasing the computational efficiency superiority of 2-Beckmann distance over Wasserstein distance.
The main contributions of the paper include:
1. Introducing the concept of p-Beckmann distance between probability measures and establishing bounds with p-Wasserstein distance.
2. Providing an exact formula for p-Beckmann distance on trees.
3. Obtaining the relationship between 2-Beckmann distance and optimal stopping/visiting time of random walks.
4. Interpreting 2-Beckmann distance as a distance of negative Sobolev seminorm and obtaining the Benamou-Brenier formula.
5. Applying 2-Beckmann distance for unsupervised learning experiments and comparing it with 2-Wasserstein distance.
The structure of the paper includes an introduction, mathematical background, general properties of p-Beckmann problem, in-depth study of 2-Beckmann problem, acknowledgments, and proofs in the appendix. The section on related work mentions previous research in the fields of optimal transportation and effective resistance, emphasizing the innovations of this paper.