Dynamic Programming on Nominal Graphs

TL;DR

This paper introduces an algebraic framework for representing graphs used in optimization problems, linking graph decompositions to dynamic programming strategies, and demonstrating applications in autonomous systems and vehicle parking optimization.

Contribution

It presents a novel algebraic specification connecting graph structures with dynamic programming, enabling flexible evaluation strategies and applications in real-world optimization problems.

Findings

Correspondence between graph decompositions and algebraic terms

Evaluation yields dynamic programming strategies independent of structure

Application to parking optimization in autonomous mobility systems

Abstract

Many optimization problems can be naturally represented as (hyper) graphs, where vertices correspond to variables and edges to tasks, whose cost depends on the values of the adjacent variables. Capitalizing on the structure of the graph, suitable dynamic programming strategies can select certain orders of evaluation of the variables which guarantee to reach both an optimal solution and a minimal size of the tables computed in the optimization process. In this paper we introduce a simple algebraic specification with parallel composition and restriction whose terms up to structural axioms are the graphs mentioned above. In addition, free (unrestricted) vertices are labelled with variables, and the specification includes operations of name permutation with finite support. We show a correspondence between the well-known tree decompositions of graphs and our terms. If an axiom of scope extension is dropped, several (hierarchical) terms actually correspond to the same graph. A suitable graphical structure can be found, corresponding to every hierarchical term. Evaluating such a graphical structure in some target algebra yields a dynamic programming strategy. If the target algebra satisfies the scope extension axiom, then the result does not depend on the particular structure, but only on the original graph. We apply our approach to the parking optimization problem developed in the ASCENS e-mobility case study, in collaboration with Volkswagen. Dynamic programming evaluations are particularly interesting for autonomic systems, where actual behavior often consists of propagating local knowledge to obtain global knowledge and getting it back for local decisions.