Papers
arxiv:2207.07696

Algorithmic Determination of the Combinatorial Structure of the Linear Regions of ReLU Neural Networks

Published on Jul 15, 2022
Authors:

Abstract

The canonical polyhedral complex decomposes ReLU network input spaces into polyhedral regions, with an algorithm determining the complete combinatorial structure using vertex locations and signs.

AI-generated summary

We algorithmically determine the regions and facets of all dimensions of the canonical polyhedral complex, the universal object into which a ReLU network decomposes its input space. We show that the locations of the vertices of the canonical polyhedral complex along with their signs with respect to layer maps determine the full facet structure across all dimensions. We present an algorithm which calculates this full combinatorial structure, making use of our theorems that the dual complex to the canonical polyhedral complex is cubical and it possesses a multiplication compatible with its facet structure. The resulting algorithm is numerically stable, polynomial time in the number of intermediate neurons, and obtains accurate information across all dimensions. This permits us to obtain, for example, the true topology of the decision boundaries of networks with low-dimensional inputs. We run empirics on such networks at initialization, finding that width alone does not increase observed topology, but width in the presence of depth does. Source code for our algorithms is accessible online at https://github.com/mmasden/canonicalpoly.

Community

Sign up or log in to comment

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2207.07696 in a model README.md to link it from this page.

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2207.07696 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2207.07696 in a Space README.md to link it from this page.

Collections including this paper 0

No Collection including this paper

Add this paper to a collection to link it from this page.