FastDOG: Fast Discrete Optimization on GPU

Abbas, Ahmed; Swoboda, Paul

doi:10.1109/cvpr52688.2022.00053

Cited by 3 publications

(29 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We now present our shape matching framework. First we will discuss improvements to (Abbas and Swoboda 2022) for solving (1) with lower runtime and better objectives. Next we propose our improved cost function which helps to obtain more accurate results by injecting more information into the optimisation problem.…”

Section: Methodsmentioning

confidence: 99%

“…In order to solve large shape matching integer linear programs (ILPs) (1), we take FastDog (Abbas and Swoboda 2022) as a starting point. Its massive parallelism and generic applicability make it a suitable choice.…”

Section: Discrete Optimisation For Shape Matchingmentioning

confidence: 99%

“…We give an overview of their approach below. The method by Abbas and Swoboda (2022) represents each constraint as a binary decision diagram (BDD) and utilises the Lagrangean dual for promoting agreement among variables in different constraints. This allows to make efficient use of GPUs, as operations in different constraints can be parallelised.…”

Section: Discrete Optimisation For Shape Matchingmentioning

confidence: 99%

“…An improved scheme for solving (2) not only helps in finding better lower bounds but also in primal recovery (upper bounds). Therefore we aim towards solving (2) more efficiently than (Abbas and Swoboda 2022) for improving both lower and upper bound.…”

Section: Discrete Optimisation For Shape Matchingmentioning

confidence: 99%

“…We are empirically able to certify global optimality in 90% of the considered problem instances. • Our solver extends the dual decomposition approach FastDog (Abbas and Swoboda 2022) by addressing the specific challenges posed by large shape matching instances. By doing so we demonstrate a 5-11 times faster convergence w.r.t.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

A Scalable Combinatorial Solver for Elastic Geometrically Consistent 3D Shape Matching

Roetzer¹,

Swoboda²,

Cremers³

et al. 2022

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

Sm-Comb FastDog Ours Maps Between Vertices ( ) Triangulation Transfer via Computed Matching Solvers Comparison on FAUST Dataset Figure 1: Left: Geometric consistency is an essential property of correspondences between shapes, meaning that neighbouring elements in M are consistently matched to neighbouring elements in N . Centre: Our proposed solution guarantees geometric consistency and thus outperforms the recent state-of-the-art shape matching approach by Cao, Roetzer, and Bernard (2023). Right: Our novel combinatorial solver is faster and finds tighter optimality gaps compared to previous approaches Sm-Comb (Roetzer et al. 2022) and FastDog (Abbas and Swoboda 2022).

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Discrete Optimisation For Shape Matchingmentioning

confidence: 99%

Section: Discrete Optimisation For Shape Matchingmentioning

confidence: 99%

Section: Discrete Optimisation For Shape Matchingmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

A Scalable Combinatorial Solver for Elastic Geometrically Consistent 3D Shape Matching

Roetzer¹,

Swoboda²,

Cremers³

et al. 2022

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

show abstract

A survey for solving mixed integer programming via machine learning

Zhang

Liu

et al. 2023

Neurocomputing

View full text Add to dashboard Cite

DOGE-Train: Discrete Optimization on GPU with End-to-End Training

Abbas,

Swoboda

2024

AAAI

View full text Add to dashboard Cite

We present a fast, scalable, data-driven approach for solving relaxations of 0-1 integer linear programs. We use a combination of graph neural networks (GNN) and a Lagrange decomposition based algorithm. We make the latter differentiable for end-to-end training and use GNNs to predict its algorithmic parameters. This allows to retain the algorithm's theoretical properties including dual feasibility and guaranteed non-decrease in the lower bound while improving it via training. We overcome suboptimal fixed points of the basic solver by additional non-parametric GNN update steps maintaining dual feasibility. For training we use an unsupervised loss. We train on smaller problems and test on larger ones showing strong generalization performance with a GNN comprising only around 10k parameters. Our solver achieves significantly faster performance and better dual objectives than its non-learned version, achieving close to optimal objective values of LP relaxations of very large structured prediction problems and on selected combinatorial ones. In particular, we achieve better objective values than specialized approximate solvers for specific problem classes while retaining their efficiency. Our solver has better any-time performance over a large time period compared to a commercial solver.

show abstract

FastDOG: Fast Discrete Optimization on GPU

Cited by 3 publications

References 37 publications

A Scalable Combinatorial Solver for Elastic Geometrically Consistent 3D Shape Matching

A Scalable Combinatorial Solver for Elastic Geometrically Consistent 3D Shape Matching

A survey for solving mixed integer programming via machine learning

DOGE-Train: Discrete Optimization on GPU with End-to-End Training

Contact Info

Product

Resources

About