Pith. sign in

Lower Bounds on the Depth of Integral ReLU Neural Networks via Lattice Polytopes

1 Pith paper cite this work. Polarity classification is still indexing.

1 Pith paper citing it
abstract

We prove that the set of functions representable by ReLU neural networks with integer weights strictly increases with the network depth while allowing arbitrary width. More precisely, we show that $\lceil\log_2(n)\rceil$ hidden layers are indeed necessary to compute the maximum of $n$ numbers, matching known upper bounds. Our results are based on the known duality between neural networks and Newton polytopes via tropical geometry. The integrality assumption implies that these Newton polytopes are lattice polytopes. Then, our depth lower bounds follow from a parity argument on the normalized volume of faces of such polytopes.

citation-role summary

background 1

citation-polarity summary

fields

math.MG 1

years

2025 1

verdicts

ACCEPT 1

roles

background 1

polarities

background 1

representative citing papers

Approximation Depth of Convex Polytopes

math.MG · 2025-07-10 · accept · novelty 8.0

For every depth d below the exact depth of the n-simplex, every depth-d polytope misses the simplex by an empty-corner distance of exactly n+1-2^d.

citing papers explorer

Showing 1 of 1 citing paper.

  • Approximation Depth of Convex Polytopes math.MG · 2025-07-10 · accept · none · ref 9 · internal anchor

    For every depth d below the exact depth of the n-simplex, every depth-d polytope misses the simplex by an empty-corner distance of exactly n+1-2^d.