Saved in:
Bibliographic Details
Main Authors: Cacciola, Matteo, Forel, Alexandre, Frangioni, Antonio, Lodi, Andrea
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2411.03535
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912106926833664
author Cacciola, Matteo
Forel, Alexandre
Frangioni, Antonio
Lodi, Andrea
author_facet Cacciola, Matteo
Forel, Alexandre
Frangioni, Antonio
Lodi, Andrea
contents Although nearly 20 years have passed since its conception, the feasibility pump algorithm remains a widely used heuristic to find feasible primal solutions to mixed-integer linear problems. Many extensions of the initial algorithm have been proposed. Yet, its core algorithm remains centered around two key steps: solving the linear relaxation of the original problem to obtain a solution that respects the constraints, and rounding it to obtain an integer solution. This paper shows that the traditional feasibility pump and many of its follow-ups can be seen as gradient-descent algorithms with specific parameters. A central aspect of this reinterpretation is observing that the traditional algorithm differentiates the solution of the linear relaxation with respect to its cost. This reinterpretation opens many opportunities for improving the performance of the original algorithm. We study how to modify the gradient-update step as well as extending its loss function. We perform extensive experiments on MIPLIB instances and show that these modifications can substantially reduce the number of iterations needed to find a solution.
format Preprint
id arxiv_https___arxiv_org_abs_2411_03535
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle The Differentiable Feasibility Pump
Cacciola, Matteo
Forel, Alexandre
Frangioni, Antonio
Lodi, Andrea
Optimization and Control
Machine Learning
Although nearly 20 years have passed since its conception, the feasibility pump algorithm remains a widely used heuristic to find feasible primal solutions to mixed-integer linear problems. Many extensions of the initial algorithm have been proposed. Yet, its core algorithm remains centered around two key steps: solving the linear relaxation of the original problem to obtain a solution that respects the constraints, and rounding it to obtain an integer solution. This paper shows that the traditional feasibility pump and many of its follow-ups can be seen as gradient-descent algorithms with specific parameters. A central aspect of this reinterpretation is observing that the traditional algorithm differentiates the solution of the linear relaxation with respect to its cost. This reinterpretation opens many opportunities for improving the performance of the original algorithm. We study how to modify the gradient-update step as well as extending its loss function. We perform extensive experiments on MIPLIB instances and show that these modifications can substantially reduce the number of iterations needed to find a solution.
title The Differentiable Feasibility Pump
topic Optimization and Control
Machine Learning
url https://arxiv.org/abs/2411.03535