Thermodynamic Algorithms for Quadratic Programming

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Bartosik, Patryk-Lipka, Donatella, Kaelan, Aifer, Maxwell, Melanson, Denis, Perarnau-Llobet, Marti, Brunner, Nicolas, Coles, Patrick J.
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912128651231232
author Bartosik, Patryk-Lipka
Donatella, Kaelan
Aifer, Maxwell
Melanson, Denis
Perarnau-Llobet, Marti
Brunner, Nicolas
Coles, Patrick J.
author_facet Bartosik, Patryk-Lipka
Donatella, Kaelan
Aifer, Maxwell
Melanson, Denis
Perarnau-Llobet, Marti
Brunner, Nicolas
Coles, Patrick J.
contents Thermodynamic computing has emerged as a promising paradigm for accelerating computation by harnessing the thermalization properties of physical systems. This work introduces a novel approach to solving quadratic programming problems using thermodynamic hardware. By incorporating a thermodynamic subroutine for solving linear systems into the interior-point method, we present a hybrid digital-analog algorithm that outperforms traditional digital algorithms in terms of speed. Notably, we achieve a polynomial asymptotic speedup compared to conventional digital approaches. Additionally, we simulate the algorithm for a support vector machine and predict substantial practical speedups with only minimal degradation in solution quality. Finally, we detail how our method can be applied to portfolio optimization and the simulation of nonlinear resistive networks.
format Preprint
id arxiv_https___arxiv_org_abs_2411_14224
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Thermodynamic Algorithms for Quadratic Programming
Bartosik, Patryk-Lipka
Donatella, Kaelan
Aifer, Maxwell
Melanson, Denis
Perarnau-Llobet, Marti
Brunner, Nicolas
Coles, Patrick J.
Emerging Technologies
Statistical Mechanics
Optimization and Control
Thermodynamic computing has emerged as a promising paradigm for accelerating computation by harnessing the thermalization properties of physical systems. This work introduces a novel approach to solving quadratic programming problems using thermodynamic hardware. By incorporating a thermodynamic subroutine for solving linear systems into the interior-point method, we present a hybrid digital-analog algorithm that outperforms traditional digital algorithms in terms of speed. Notably, we achieve a polynomial asymptotic speedup compared to conventional digital approaches. Additionally, we simulate the algorithm for a support vector machine and predict substantial practical speedups with only minimal degradation in solution quality. Finally, we detail how our method can be applied to portfolio optimization and the simulation of nonlinear resistive networks.
title Thermodynamic Algorithms for Quadratic Programming
topic Emerging Technologies
Statistical Mechanics
Optimization and Control
url https://arxiv.org/abs/2411.14224