Are Graph Neural Networks Optimal Approximation Algorithms?
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2023
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866917794570829824 |
|---|---|
| author | Yau, Morris Karalias, Nikolaos Lu, Eric Xu, Jessica Jegelka, Stefanie |
| author_facet | Yau, Morris Karalias, Nikolaos Lu, Eric Xu, Jessica Jegelka, Stefanie |
| contents | In this work we design graph neural network architectures that capture optimal approximation algorithms for a large class of combinatorial optimization problems, using powerful algorithmic tools from semidefinite programming (SDP). Concretely, we prove that polynomial-sized message-passing algorithms can represent the most powerful polynomial time algorithms for Max Constraint Satisfaction Problems assuming the Unique Games Conjecture. We leverage this result to construct efficient graph neural network architectures, OptGNN, that obtain high-quality approximate solutions on landmark combinatorial optimization problems such as Max-Cut, Min-Vertex-Cover, and Max-3-SAT. Our approach achieves strong empirical results across a wide range of real-world and synthetic datasets against solvers and neural baselines. Finally, we take advantage of OptGNN's ability to capture convex relaxations to design an algorithm for producing bounds on the optimal solution from the learned embeddings of OptGNN. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2310_00526 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Are Graph Neural Networks Optimal Approximation Algorithms? Yau, Morris Karalias, Nikolaos Lu, Eric Xu, Jessica Jegelka, Stefanie Machine Learning Artificial Intelligence Discrete Mathematics Data Structures and Algorithms In this work we design graph neural network architectures that capture optimal approximation algorithms for a large class of combinatorial optimization problems, using powerful algorithmic tools from semidefinite programming (SDP). Concretely, we prove that polynomial-sized message-passing algorithms can represent the most powerful polynomial time algorithms for Max Constraint Satisfaction Problems assuming the Unique Games Conjecture. We leverage this result to construct efficient graph neural network architectures, OptGNN, that obtain high-quality approximate solutions on landmark combinatorial optimization problems such as Max-Cut, Min-Vertex-Cover, and Max-3-SAT. Our approach achieves strong empirical results across a wide range of real-world and synthetic datasets against solvers and neural baselines. Finally, we take advantage of OptGNN's ability to capture convex relaxations to design an algorithm for producing bounds on the optimal solution from the learned embeddings of OptGNN. |
| title | Are Graph Neural Networks Optimal Approximation Algorithms? |
| topic | Machine Learning Artificial Intelligence Discrete Mathematics Data Structures and Algorithms |
| url | https://arxiv.org/abs/2310.00526 |