Abstract:Covering numbers are a powerful tool used in the development of approximation algorithms, randomized dimension reduction methods, smoothed complexity analysis, and others. In this paper we prove upper bounds on the covering number of numerous sets in Euclidean space, namely real algebraic varieties, images of polynomial maps and semialgebraic sets in terms of the number of variables and degrees of the polynomials involved. The bounds remarkably improve the best known general bound by Yomdin-Comte, and our proof is much more straightforward. In particular, our result gives new bounds on the volume of the tubular neighborhood of the image of a polynomial map and a semialgebraic set, where results for varieties by Lotz and Basu-Lerario are not directly applicable. We illustrate the power of the result on three computational applications. Firstly, we derive a near-optimal bound on the covering number of low rank CP tensors, quantifying their approximation properties and filling in an important missing piece of theory for tensor dimension reduction and reconstruction. Secondly, we prove a bound on the required dimension for the randomized sketching of polynomial optimization problems, which controls how much computation can be saved through randomization without sacrificing solution quality. Finally, we deduce generalization error bounds for deep neural networks with rational or ReLU activation functions, improving or matching the best known results in the machine learning literature while helping to quantify the impact of architecture choice on generalization error.

An improvement upon unmixed decomposition of an algebraic variety

Computing with Semi-Algebraic Sets Represented by Triangular Decomposition

Generic Regular Decompositions for Generic Zero-Dimensional Systems

A Geometric Approach to Cylindrical Algebraic Decomposition

Stability of Triangular Decomposition and Comprehensive Triangular Decomposition

Computing Cylindrical Algebraic Decomposition Via Triangular Decomposition

Generic Regular Decompositions for Parametric Polynomial Systems

Square-Free Pure Triangular Decomposition of Zero-Dimensional Polynomial Systems

Triangular Decomposition of Semi-algebraic Systems

Decomposing abelian varieties into simple factors: algorithms and applications

Zero Decomposition with Multiplicity of Zero-Dimensional Polynomial Systems

Sparse Tensors and Subdivision Methods for Finding the Zero Set of Polynomial Equations

An Improvement over the GVW Algorithm for Inhomogeneous Polynomial Systems.

Decomposing Solution Sets of Polynomial Systems: A New Parallel Monodromy Breakup Algorithm

Fine multidegrees, universal Grobner bases, and matrix Schubert varieties

The decomposition theorem and the intersection cohomology of quotients in algebraic geometry

Covering Number of Real Algebraic Varieties and Beyond: Improved Bounds and Applications

Symmetric decompositions, triangulations and real-rootedness

On The Group Algebra Decomposition of a Jacobian Variety

A Syzygial Method for Equidimensional Decomposition

Symmetric Tensor Decompositions On Varieties