Towards Lower Bounds on the Depth of ReLU Neural Networks

Authors: Christoph Hertrich, Amitabh Basu, Marco Di Summa, Martin Skutella

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical Using techniques from mixed-integer optimization, polyhedral theory, and tropical geometry, we provide a mathematical counterbalance to the universal approximation theorems... and In Section 2, we resolve Conjecture 1.2 for k = 2, under a natural assumption on the breakpoints of the function represented by any intermediate neuron. We achieve this result by leveraging techniques from mixed-integer programming to analyze the set of functions computable by certain NNs. This provides a computational proof of Theorem 2.5.
Researcher Affiliation Academia Christoph Hertrich Technische Universität Berlin Berlin, Germany christoph.hertrich@posteo.de, Amitabh Basu Johns Hopkins University Baltimore, USA basu.amitabh@jhu.edu, Marco Di Summa Università degli Studi di Padova Padua, Italy disumma@math.unipd.it, Martin Skutella Technische Universität Berlin Berlin, Germany martin.skutella@tu-berlin.de
Pseudocode No No structured pseudocode or algorithm blocks were found in the paper.
Open Source Code No The paper does not provide an explicit statement or link for the open-source code for the methodology described.
Open Datasets No This is a theoretical paper that does not involve experimental training on datasets.
Dataset Splits No This is a theoretical paper and does not involve dataset splits for training, validation, or testing.
Hardware Specification No The paper mentions a 'computer-aided proof' and solving a Mixed-Integer Program (MIP) but does not provide specific hardware details (e.g., CPU, GPU models, or cloud computing specifications) used for these computations.
Software Dependencies Yes Gurobi Optimization, LLC. Gurobi optimizer reference manual, 2021. and The Sage Developers. Sage Math, the Sage Mathematics Software System (Version 9.0), 2020.
Experiment Setup No This is a theoretical paper that performs a computer-aided proof involving a Mixed-Integer Program (MIP), but it does not describe specific experimental setup details such as hyperparameters or training configurations typically found in empirical machine learning studies.