Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

The Computational Complexity of Counting Linear Regions in ReLU Neural Networks

Authors: Moritz Stargalla, Christoph Hertrich, Daniel Reichman

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical Our paper is of theoretical nature and we strive towards a thorough understanding of the problem of counting regions from a computational complexity perspective.
Researcher Affiliation Academia Moritz Stargalla University of Technology Nuremberg EMAIL Christoph Hertrich University of Technology Nuremberg EMAIL Daniel Reichman Worcester Polytechnic Institute EMAIL
Pseudocode Yes Algorithm 1 SEARCHAFFINEPIECE Input: A Re LU network N and a vector (a1, . . . , an, b) Qn+1. Output: 1 if Pn i=1 aixi + b is a function of an affine region of N, else 0. 1: for a {0, 1}s(N) do 2: if dim Sa = n then (Lemma A.5) 3: if Pn i=1 aixi + b = f a N(x) then return 1 (Lemma A.4) return 0 Algorithm 2 EXHAUSTIVESEARCH Input: A Re LU network N. Output: Number of affine regions of N. 1: nmax = max{n0, n1, . . . , nd+1} 2: U = 236d2n2 max Amax (Lemma A.3) 3: R = 0 4: for (a, b) { U, . . . , U}n+1 {1, . . . , U}n+1 do 5: if gcd(ai, bi) = 1 for i [n + 1] then 6: R R + SEARCHAFFINEPIECE(N, ( a1 b1 , . . . , an+1 bn+1 )) (Lemma B.4) return R
Open Source Code No Justification: No experiments.
Open Datasets No Justification: No experiments.
Dataset Splits No Justification: No experiments.
Hardware Specification No Justification: No experiments.
Software Dependencies No Justification: No experiments.
Experiment Setup No Justification: No experiments.