Tight Bounds on Low-degree Spectral Concentration of Submodular and XOS functions.

Sub modular and fractionally sub additive (or equivalently XOS) functions play a fundamental role in combinatorial optimization, algorithmic game theory and machine learning. Motivated by learn ability of these classes of functions from random examples, we consider the question of how well such functions can be approximated by low-degree polynomials in L2 norm over the uniform distribution. This question is equivalent to understanding of the concentration of Fourier weight on low-degree coefficients, a central concept in Fourier analysis. We show that. For any sub modular function f:{0, 1}n → [0, 1], deg l2ε(f) = O(log(1/ε)/ε4/5) there is a sub modular function that requires degree Ω(1/ε4/5}). For any XOS function f:{0, 1}n → [0, 1], deg l2ε(f) = O(1/ε) and there exists an XOS function that requires degree Ω(1/ε). This improves on previous approaches that all showed an upper bound of O(1/ε2) for sub modular and XOS functions. The best previous lower bound was Ω(1/ε{2/3}) for monotone sub modular functions. Our techniques reveal new structural properties of sub modular and XOS functions and the upper bounds lead to nearly optimal PAC learning algorithms for these classes of functions.