Convexity Certificates from Hessians

Authors: Julien Klaus, Niklas Merk, Konstantin Wiedom, Sören Laue, Joachim Giesen

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical Here, we implement this approach for a class of functions that is rich enough to support classical machine learning. For this class of functions it was recently shown how to compute computational graphs of their Hessians. We show how to check these graphs for positive semidefiniteness. We compare our implementation of the Hessian approach with the well-established disciplined convex programming (DCP) approach and prove that the Hessian approach is at least as powerful as the DCP approach for differentiable functions.
Researcher Affiliation Collaboration Julien Klaus Friedrich Schiller University Jena julien.klaus@uni-jena.de Niklas Merk Friedrich Schiller University Jena niklas.merk@uni-jena.de Konstantin Wiedom Friedrich Schiller University Jena konstantin.wiedom@uni-jena.de Sören Laue Technical University of Kaiserslautern Data Assessment Solutions Gmb H laue@cs.uni-kl.de Joachim Giesen Friedrich Schiller University Jena joachim.giesen@uni-jena.de
Pseudocode Yes Algorithm 1 Certify convexity of a Hessian; Algorithm 2 Compute the positivity interval for a node v
Open Source Code No The paper states '[N/A]' for the question 'Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)?' in its ethics checklist, and no explicit link or statement of code availability is found elsewhere.
Open Datasets No The paper focuses on theoretical development and algorithmic implementation for certifying convexity, not on empirical evaluation involving training on datasets. Therefore, no dataset access information is provided.
Dataset Splits No The paper does not report empirical experiments with dataset splits, as its focus is on theoretical and algorithmic contributions.
Hardware Specification No The paper indicates '[N/A]' for hardware specifications in its ethics checklist (3d: 'Did you include the total amount of compute and the type of resources used...'), consistent with its theoretical focus.
Software Dependencies No The paper discusses various software frameworks (e.g., CVX, GENO) in the context of related work or comparison, but it does not specify any particular software dependencies with version numbers required to reproduce its own implementation.
Experiment Setup No The paper describes a theoretical framework and algorithmic implementation but does not provide specific experimental setup details such as hyperparameters or training configurations, as it does not report empirical experiments.