Differentiable Extensions with Rounding Guarantees for Combinatorial Optimization over Permutations

Robert R. Nerem, Zhishang Luo, Akbar Rafiey, Yusu Wang

Abstract: We present Birkhoff Extension (BE), an almost-everywhere-differentiable continuous polytime-computable extension of any real-valued function on permutations to doubly stochastic matrices. Our approach is based on Birkhoff decomposition (also referred to as Birkhoff von-Neumann decomposition) which allows construction of an extension that is always a convex combination of the objective's values at permutations. We show how to construct a specific family of Birkhoff decompositions that are continuous. In addition to continuity, our extension has several nice properties making it appealing for optimization problems. First, BE provides a rounding guarantee, namely any solution to the extension can be efficiently rounded to a permutation without increasing the function value. Furthermore, an approximate solution in the relaxed case (with extension) will give rise to an approximate solution in the space of permutations. Second, using BE, any real-valued optimization objective on permutations can be extended to an almost everywhere differentiable objective function over the space of doubly stochastic matrices. This makes our BE amenable to not only gradient-descent based optimizations, but also unsupervised neural combinatorial optimization where training often requires a differentiable loss. Third, based on the above properties, we present a simple optimization procedure which can be readily combined with existing optimization approaches to offer local improvements (i.e., the quality of the final solution is no worse than the initial solution). We present preliminary experimental results to verify our theoretical results on several combinatorial optimization problems related to permutations.