Adaptive Proximal Gradient Method for Convex Optimization

Authors: Yura Malitsky, Konstantin Mishchenko

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Section 6 (see also Appendix D), we conduct experiments to evaluate the proposed method against different linesearch variants.
Researcher Affiliation Collaboration Yura Malitsky Faculty of Mathematics University of Vienna, Austria yurii.malitskyi@univie.ac.at Konstantin Mishchenko Samsung AI Center, UK konsta.mish@gmail.com
Pseudocode Yes Algorithm 1 Adaptive gradient descent; Algorithm 2 Adaptive gradient descent-2; Algorithm 3 Adaptive proximal gradient method
Open Source Code Yes 2https://github.com/ymalitsky/Ad Prox GD
Open Datasets No The paper generates synthetic data for its experiments (e.g., 'We generated a random y Rn', 'We created matrix A by multiplying matrices U and V') rather than using existing public datasets with specific access information.
Dataset Splits No The paper describes how data was generated for various problems but does not specify training, validation, or test dataset splits in terms of percentages or sample counts for reproduction, nor does it reference standard predefined splits.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., library names, framework versions) needed to replicate the experiments.
Experiment Setup Yes An efficient implementation of Armijo s linesearch requires two parameters, s > 1 and r < 1. In the k-th iteration, the first iteration of linesearch starts from αk = sαk 1... The choice of (s, r) matters a lot. For Maximum likelihood estimate problem: n = 100, l = 0.1, u = 10, M = 50. For Low-rank matrix completion: n = 100, r = 20. For Minimal length piecewise-linear curve: m = 50, n = 200.