Transfer Learning from Foundational Optimization Embeddings to Unsupervised SAT Representations

Koyena Pal; Serdar Kadioglu

arxiv: 2604.15448 · v1 · submitted 2026-04-16 · 💻 cs.LG · cs.AI· cs.LO

Transfer Learning from Foundational Optimization Embeddings to Unsupervised SAT Representations

Koyena Pal , Serdar Kadioglu This is my paper

Pith reviewed 2026-05-10 11:57 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.LO

keywords transfer learningSATembeddingsunsupervised learningbipartite graphsconstraint satisfactionmixed-integer programmingfoundational models

0 comments

The pith

Pre-trained embeddings from optimization problems apply directly to SAT instances for unsupervised tasks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that embeddings learned on mixed-integer programming can be reused on Boolean satisfiability by converting CNF formulas into the same bipartite constraint-variable graph format. This mapping lets the existing model extract structural patterns in SAT problems without any architectural changes or new training. The embeddings then support unsupervised work such as grouping similar instances and recognizing different SAT distributions. A sympathetic reader would care because it opens a route to shared representations across optimization and decision problems, lowering the need for domain-specific labeled data.

Core claim

Foundational optimization embeddings generalize to decision problems such as SAT by mapping CNF formulas into the bipartite constraint-variable graph representation already used for MIPs. This allows direct reuse of the pre-trained embedding model without changes or supervised fine-tuning. The embeddings capture structural regularities in SAT instances and enable unsupervised tasks including instance clustering and distribution identification, providing the first demonstration that such transfer from optimization to constraint satisfaction is possible.

What carries the argument

The bipartite constraint-variable graph representation, which encodes both MIPs and SAT formulas uniformly so the same pre-trained embedding model can be applied without adaptation.

If this is right

SAT instances can be clustered by structural features learned from the embeddings.
Distributions of SAT problems can be identified in an unsupervised manner.
A unified representational framework for optimization and decision problems becomes feasible.
Dependence on solver-generated labels for SAT-related tasks is reduced.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same embeddings may transfer to other constraint satisfaction problems such as general CSPs or scheduling tasks.
Instance features from these embeddings could help SAT solvers choose heuristics more quickly.
The approach invites tests on whether the embeddings predict SAT hardness or locate phase-transition thresholds.
Graph-based learning methods for combinatorial problems may benefit from this cross-domain reuse pattern.

Load-bearing premise

The bipartite graph format created for MIPs already preserves the structural features that matter for SAT, so no adaptation or loss occurs when reusing the model.

What would settle it

The embeddings produce clustering of SAT instances no better than random vectors or fail to separate instances drawn from known different distributions when measured by standard unsupervised metrics.

read the original abstract

Foundational optimization embeddings have recently emerged as powerful pre-trained representations for mixed-integer programming (MIP) problems. These embeddings were shown to enable cross-domain transfer and reduce reliance on solver-generated labels. In this work, we investigate whether such representations generalize beyond optimization to decision problems, focusing on Boolean satisfiability (SAT). We adapt the foundational optimization architecture to SAT by mapping CNF formulas into the same bipartite constraint-variable graph representation used for MIPs. This allows direct reuse of the pre-trained embedding model without architectural changes or supervised fine-tuning. Our results show that these embeddings capture structural regularities in SAT instances and support unsupervised tasks such as instance clustering and distribution identification. We demonstrate, for the first time, that foundational optimization embeddings can transfer to constraint satisfaction domains. Our findings is a step toward a unified representational framework for both optimization and decision problems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

MIP embeddings reuse on SAT via bipartite graphs is a clean first try, but the results may reflect generic graph features more than transferred optimization knowledge.

read the letter

The key takeaway is that the authors map SAT problems to the bipartite variable-constraint graph format from their earlier MIP work and apply the same pre-trained embedding model directly. They find that the resulting representations support unsupervised clustering of SAT instances and help distinguish different distributions of problems. This is new because it extends the embedding approach beyond optimization into pure decision problems like SAT, and they position it as the first such demonstration. The fact that no changes to the model or supervised fine-tuning are needed is a practical plus. It keeps things simple and shows the embeddings have some cross-domain utility out of the box. The paper does a reasonable job of describing the adaptation step, which is mostly about how to represent CNF formulas in the graph without adding weights or other MIP-specific elements. Where it gets softer is on the interpretation of the results. The stress-test concern is worth taking seriously: SAT clauses are unweighted disjunctions, while MIP constraints involve coefficients and bounds. If the pre-trained encoder relies on those numeric features, then on SAT graphs it might only be learning basic connectivity patterns that any graph model could pick up. Without ablations comparing to embeddings trained from scratch on SAT or to non-optimization graph methods, it's difficult to attribute the clustering performance specifically to the foundational optimization knowledge. The abstract is light on details, so the full paper presumably includes the actual metrics, baselines, and statistical significance. Those need to be convincing for the transfer to be taken as more than a proof of concept. Overall, this paper is aimed at researchers working on machine learning for combinatorial optimization and constraint satisfaction. It could be useful as an early exploration of unified embeddings, but readers will want to see robust evidence that the transfer is meaningful rather than superficial. I think it deserves peer review to sort out the experimental questions and see if the claims hold under scrutiny.

Referee Report

2 major / 1 minor

Summary. The manuscript claims that foundational optimization embeddings pre-trained on MIP problems can be transferred to SAT by mapping CNF formulas to the identical bipartite constraint-variable graph representation used for MIPs. This mapping permits direct reuse of the pre-trained model with no architectural changes or supervised fine-tuning. The embeddings are asserted to capture structural regularities in SAT instances that enable unsupervised tasks such as instance clustering and distribution identification, constituting the first demonstration of transfer from optimization embeddings to constraint satisfaction domains and a step toward a unified representational framework.

Significance. If the transfer results hold under rigorous evaluation, the work would offer a meaningful contribution to cross-domain representation learning in combinatorial optimization and decision problems. The direct-reuse strategy without fine-tuning is efficient and could reduce reliance on solver-generated labels. However, the significance is currently difficult to gauge because the abstract supplies no quantitative metrics, baselines, or experimental details to confirm that the captured regularities are optimization-derived rather than generic graph properties.

major comments (2)

[Abstract] Abstract: The abstract asserts that 'our results show that these embeddings capture structural regularities in SAT instances and support unsupervised tasks such as instance clustering and distribution identification' yet supplies no metrics, baselines, dataset sizes, clustering algorithms, or evaluation protocols. This absence prevents verification that the data support the transfer claims and leaves the central empirical assertion unassessable.
[Graph mapping description] Graph mapping description: The load-bearing assumption that the bipartite variable-clause graph developed for MIP linear inequalities can be reused unchanged for SAT disjunctive clauses preserves optimization-derived features is not tested. MIP constraints contain coefficients and bounds absent from SAT; if the pre-trained encoder is sensitive to these, the SAT embeddings may reflect only generic topology. The manuscript should include an ablation (e.g., pre-trained vs. randomly initialized encoder on identical SAT graphs) or embedding visualizations to establish that the features are indeed transferred optimization knowledge rather than generic graph statistics.

minor comments (1)

[Abstract] Abstract: Grammatical error in the final sentence: 'Our findings is a step' should read 'Our findings are a step'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight opportunities to strengthen the empirical presentation and validation of our transfer claims. We address each major comment below and will incorporate the suggested improvements in the revised manuscript.

read point-by-point responses

Referee: [Abstract] Abstract: The abstract asserts that 'our results show that these embeddings capture structural regularities in SAT instances and support unsupervised tasks such as instance clustering and distribution identification' yet supplies no metrics, baselines, dataset sizes, clustering algorithms, or evaluation protocols. This absence prevents verification that the data support the transfer claims and leaves the central empirical assertion unassessable.

Authors: We agree that the abstract would benefit from additional quantitative context to make the claims immediately verifiable. In the revision we will expand the abstract to report key details including the number of SAT instances drawn from standard benchmarks (e.g., SATLIB and random 3-SAT distributions), the clustering method (k-means with silhouette coefficient), and comparative metrics against baselines such as random graph embeddings and untrained GNNs. These additions will allow readers to assess the strength of the unsupervised transfer results without altering the overall length or focus of the abstract. revision: yes
Referee: [Graph mapping description] Graph mapping description: The load-bearing assumption that the bipartite variable-clause graph developed for MIP linear inequalities can be reused unchanged for SAT disjunctive clauses preserves optimization-derived features is not tested. MIP constraints contain coefficients and bounds absent from SAT; if the pre-trained encoder is sensitive to these, the SAT embeddings may reflect only generic topology. The manuscript should include an ablation (e.g., pre-trained vs. randomly initialized encoder on identical SAT graphs) or embedding visualizations to establish that the features are indeed transferred optimization knowledge rather than generic graph statistics.

Authors: The referee correctly identifies that an explicit test is needed to confirm the embeddings reflect optimization-derived structure rather than generic bipartite-graph properties. While the direct transfer without fine-tuning already produces meaningful unsupervised clustering and distribution separation on SAT instances (outcomes unlikely under purely generic topology), we will add the requested ablation: performance of the pre-trained encoder versus a randomly initialized encoder on identical SAT graphs. We will also include t-SNE visualizations of the resulting embeddings to illustrate how different SAT distributions separate in the learned space. These additions will be placed in the experimental section and will directly address the concern. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical transfer via graph mapping is self-contained

full rationale

The paper presents an empirical transfer of a pre-trained MIP embedding model to SAT instances by reusing an existing bipartite graph representation without architectural modification or supervised fine-tuning. No equations, predictions, or first-principles derivations are claimed; results rest on experimental clustering and distribution identification on external SAT data. The mapping step is a direct reuse of prior architecture rather than a self-referential fit or ansatz, and no self-citation load-bearing chains reduce the central claim to its own inputs. This is a standard non-circular empirical study.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; the method relies on an existing pre-trained model and a direct graph isomorphism assumption between MIP and SAT representations.

pith-pipeline@v0.9.0 · 5445 in / 897 out tokens · 25661 ms · 2026-05-10T11:57:50.941701+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

[1]

On the glucose sat solver

Gilles Audemard and Laurent Simon. On the glucose sat solver. International Journal on Artificial Intelligence Tools , 27(01):1840001, 2018. https://arxiv.org/abs/https://doi.org/10.1142/S0218213018400018 arXiv:https://doi.org/10.1142/S0218213018400018 , https://doi.org/10.1142/S0218213018400018 doi:10.1142/S0218213018400018

work page doi:10.1142/s0218213018400018 2018
[2]

Predicting propositional satisfiability via end-to-end learning

Chris Cameron, Rex Chen, Jason Hartford, and Kevin Leyton-Brown. Predicting propositional satisfiability via end-to-end learning. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 34, pages 3324--3331, 2020

work page 2020
[3]

MILP-SAT-GNN: Yet another neural SAT solver.arXiv preprint arXiv:2507.01825,

Franco Alberto Cardillo, Hamza Khyari, and Umberto Straccia. Milp-sat-gnn: Yet another neural sat solver, 2025. URL: https://arxiv.org/abs/2507.01825, https://arxiv.org/abs/2507.01825 arXiv:2507.01825

work page arXiv 2025
[4]

Enhancing modern sat solver with machine learning method

Guanting Chen and Jia Wang. Enhancing modern sat solver with machine learning method. In Proceedings of the Great Lakes Symposium on VLSI 2025 , GLSVLSI '25, page 886–892, New York, NY, USA, 2025. Association for Computing Machinery. https://doi.org/10.1145/3716368.3735251 doi:10.1145/3716368.3735251

work page doi:10.1145/3716368.3735251 2025
[5]

Augment with care: Contrastive learning for combinatorial problems

Haonan Duan, Pashootan Vaezipoor, Max B Paulus, Yangjun Ruan, and Chris Maddison. Augment with care: Contrastive learning for combinatorial problems. In International Conference on Machine Learning , pages 5627--5642. PMLR, 2022

work page 2022
[6]

Exact combinatorial optimization with graph convolutional neural networks

Maxime Gasse, Didier Ch \'e telat, Nicola Ferroni, Laurent Charlin, and Andrea Lodi. Exact combinatorial optimization with graph convolutional neural networks. Neural Information Processing Systems , 32:15554--15566, 2019

work page 2019
[7]

A modularity-based random sat instances generator

Jes \'u s Gir \'a ldez-Cru and Jordi Levy. A modularity-based random sat instances generator. 2015

work page 2015
[8]

Locality in random sat instances

Jes \'u s Gir \'a ldez-Cru and Jordi Levy. Locality in random sat instances. In IJCAI , volume 17, pages 638--644, 2017

work page 2017
[9]

Hamilton, Zhitao Ying, and Jure Leskovec

William L. Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs. In Neural Information Processing Systems , pages 1024--1034, 2017

work page 2017
[10]

Enhancing sat solvers with glue variable predictions, 2020

Jesse Michael Han. Enhancing sat solvers with glue variable predictions, 2020. URL: https://arxiv.org/abs/2007.02559, https://arxiv.org/abs/2007.02559 arXiv:2007.02559

work page arXiv 2020
[11]

Can q-learning with graph networks learn a generalizable branching heuristic for a sat solver? In Advances in Neural Information Processing Systems 32 , 2020

Vitaly Kurin, Saad Godil, Shimon Whiteson, and Bryan Catanzaro. Can q-learning with graph networks learn a generalizable branching heuristic for a sat solver? In Advances in Neural Information Processing Systems 32 , 2020

work page 2020
[12]

Cnfgen: A generator of crafted benchmarks

Massimo Lauria, Jan Elffers, Jakob Nordstr \"o m, and Marc Vinyals. Cnfgen: A generator of crafted benchmarks. In International Conference on Theory and Applications of Satisfiability Testing , pages 464--473. Springer, 2017

work page 2017
[13]

G4 SATB ench: Benchmarking and advancing SAT solving with graph neural networks

Zhaoyu Li, Jinpei Guo, and Xujie Si. G4 SATB ench: Benchmarking and advancing SAT solving with graph neural networks. Transactions on Machine Learning Research , 2024. URL: https://openreview.net/forum?id=7VB5db72lr

work page 2024
[14]

Conflict-driven clause learning sat solvers

Joao Marques-Silva, In \^e s Lynce, and Sharad Malik. Conflict-driven clause learning sat solvers. Handbook of satisfiability , pages 131--153, 2009

work page 2009
[15]

Guiding high-performance sat solvers with unsat-core predictions

Daniel Selsam and Nikolaj Bj rner. Guiding high-performance sat solvers with unsat-core predictions. In Mikol \'a s Janota and In \^e s Lynce, editors, Theory and Applications of Satisfiability Testing -- SAT 2019 , pages 336--353, Cham, 2019. Springer International Publishing

work page 2019
[16]

Daniel Selsam, Matthew Lamm, Benedikt B\" u nz, Percy Liang, Leonardo de Moura, and David L. Dill. Learning a SAT solver from single-bit supervision. In International Conference on Learning Representations , 2019. URL: https://openreview.net/forum?id=HJMC_iA5tm

work page 2019
[17]

Forge: Foundational optimization representations from graph embeddings, 2025

Zohair Shafi and Serdar Kadioglu. Forge: Foundational optimization representations from graph embeddings, 2025. URL: https://arxiv.org/abs/2508.20330, https://arxiv.org/abs/2508.20330 arXiv:2508.20330

work page arXiv 2025
[18]

Neuroback: Improving CDCL SAT solving using graph neural networks

Wenxi Wang, Yang Hu, Mohit Tiwari, Sarfraz Khurshid, Kenneth McMillan, and Risto Miikkulainen. Neuroback: Improving CDCL SAT solving using graph neural networks. In The Twelfth International Conference on Learning Representations , 2024

work page 2024
[19]

Understanding how dimension reduction tools work: an empirical approach to deciphering t-sne, umap, trimap, and pacmap for data visualization

Yingfan Wang, Haiyang Huang, Cynthia Rudin, and Yaron Shaposhnik. Understanding how dimension reduction tools work: an empirical approach to deciphering t-sne, umap, trimap, and pacmap for data visualization. Journal of Machine Learning Research , 22(201):1--73, 2021

work page 2021
[20]

Vqgraph: Rethinking graph representation space for bridging gnns and mlps

Ling Yang, Ye Tian, Minkai Xu, Zhongyi Liu, Shenda Hong, Wei Qu, Wentao Zhang, Bin Cui, Muhan Zhang, and Jure Leskovec. Vqgraph: Rethinking graph representation space for bridging gnns and mlps. In International Conference on Learning Representations , 2024

work page 2024
[21]

Nlocalsat: Boosting local search with solution prediction

Wenjie Zhang, Zeyu Sun, Qihao Zhu, Ge Li, Shaowei Cai, Yingfei Xiong, and Lu Zhang. Nlocalsat: Boosting local search with solution prediction. In Christian Bessiere, editor, Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI-20 , pages 1177--1183. International Joint Conferences on Artificial Intelligence Orga...

work page doi:10.24963/ijcai.2020/164 2020
[22]

Grass: Combining graph neural networks with expert knowledge for sat solver selection

Zhanguang Zhang, Didier Ch \'e telat, Joseph Cotnareanu, Amur Ghose, Wenyi Xiao, Hui-Ling Zhen, Yingxue Zhang, Jianye Hao, Mark Coates, and Mingxuan Yuan. Grass: Combining graph neural networks with expert knowledge for sat solver selection. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages 6301--6311, 2024

work page 2024

[1] [1]

On the glucose sat solver

Gilles Audemard and Laurent Simon. On the glucose sat solver. International Journal on Artificial Intelligence Tools , 27(01):1840001, 2018. https://arxiv.org/abs/https://doi.org/10.1142/S0218213018400018 arXiv:https://doi.org/10.1142/S0218213018400018 , https://doi.org/10.1142/S0218213018400018 doi:10.1142/S0218213018400018

work page doi:10.1142/s0218213018400018 2018

[2] [2]

Predicting propositional satisfiability via end-to-end learning

Chris Cameron, Rex Chen, Jason Hartford, and Kevin Leyton-Brown. Predicting propositional satisfiability via end-to-end learning. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 34, pages 3324--3331, 2020

work page 2020

[3] [3]

MILP-SAT-GNN: Yet another neural SAT solver.arXiv preprint arXiv:2507.01825,

Franco Alberto Cardillo, Hamza Khyari, and Umberto Straccia. Milp-sat-gnn: Yet another neural sat solver, 2025. URL: https://arxiv.org/abs/2507.01825, https://arxiv.org/abs/2507.01825 arXiv:2507.01825

work page arXiv 2025

[4] [4]

Enhancing modern sat solver with machine learning method

Guanting Chen and Jia Wang. Enhancing modern sat solver with machine learning method. In Proceedings of the Great Lakes Symposium on VLSI 2025 , GLSVLSI '25, page 886–892, New York, NY, USA, 2025. Association for Computing Machinery. https://doi.org/10.1145/3716368.3735251 doi:10.1145/3716368.3735251

work page doi:10.1145/3716368.3735251 2025

[5] [5]

Augment with care: Contrastive learning for combinatorial problems

Haonan Duan, Pashootan Vaezipoor, Max B Paulus, Yangjun Ruan, and Chris Maddison. Augment with care: Contrastive learning for combinatorial problems. In International Conference on Machine Learning , pages 5627--5642. PMLR, 2022

work page 2022

[6] [6]

Exact combinatorial optimization with graph convolutional neural networks

Maxime Gasse, Didier Ch \'e telat, Nicola Ferroni, Laurent Charlin, and Andrea Lodi. Exact combinatorial optimization with graph convolutional neural networks. Neural Information Processing Systems , 32:15554--15566, 2019

work page 2019

[7] [7]

A modularity-based random sat instances generator

Jes \'u s Gir \'a ldez-Cru and Jordi Levy. A modularity-based random sat instances generator. 2015

work page 2015

[8] [8]

Locality in random sat instances

Jes \'u s Gir \'a ldez-Cru and Jordi Levy. Locality in random sat instances. In IJCAI , volume 17, pages 638--644, 2017

work page 2017

[9] [9]

Hamilton, Zhitao Ying, and Jure Leskovec

William L. Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs. In Neural Information Processing Systems , pages 1024--1034, 2017

work page 2017

[10] [10]

Enhancing sat solvers with glue variable predictions, 2020

Jesse Michael Han. Enhancing sat solvers with glue variable predictions, 2020. URL: https://arxiv.org/abs/2007.02559, https://arxiv.org/abs/2007.02559 arXiv:2007.02559

work page arXiv 2020

[11] [11]

Can q-learning with graph networks learn a generalizable branching heuristic for a sat solver? In Advances in Neural Information Processing Systems 32 , 2020

Vitaly Kurin, Saad Godil, Shimon Whiteson, and Bryan Catanzaro. Can q-learning with graph networks learn a generalizable branching heuristic for a sat solver? In Advances in Neural Information Processing Systems 32 , 2020

work page 2020

[12] [12]

Cnfgen: A generator of crafted benchmarks

Massimo Lauria, Jan Elffers, Jakob Nordstr \"o m, and Marc Vinyals. Cnfgen: A generator of crafted benchmarks. In International Conference on Theory and Applications of Satisfiability Testing , pages 464--473. Springer, 2017

work page 2017

[13] [13]

G4 SATB ench: Benchmarking and advancing SAT solving with graph neural networks

Zhaoyu Li, Jinpei Guo, and Xujie Si. G4 SATB ench: Benchmarking and advancing SAT solving with graph neural networks. Transactions on Machine Learning Research , 2024. URL: https://openreview.net/forum?id=7VB5db72lr

work page 2024

[14] [14]

Conflict-driven clause learning sat solvers

Joao Marques-Silva, In \^e s Lynce, and Sharad Malik. Conflict-driven clause learning sat solvers. Handbook of satisfiability , pages 131--153, 2009

work page 2009

[15] [15]

Guiding high-performance sat solvers with unsat-core predictions

Daniel Selsam and Nikolaj Bj rner. Guiding high-performance sat solvers with unsat-core predictions. In Mikol \'a s Janota and In \^e s Lynce, editors, Theory and Applications of Satisfiability Testing -- SAT 2019 , pages 336--353, Cham, 2019. Springer International Publishing

work page 2019

[16] [16]

Daniel Selsam, Matthew Lamm, Benedikt B\" u nz, Percy Liang, Leonardo de Moura, and David L. Dill. Learning a SAT solver from single-bit supervision. In International Conference on Learning Representations , 2019. URL: https://openreview.net/forum?id=HJMC_iA5tm

work page 2019

[17] [17]

Forge: Foundational optimization representations from graph embeddings, 2025

Zohair Shafi and Serdar Kadioglu. Forge: Foundational optimization representations from graph embeddings, 2025. URL: https://arxiv.org/abs/2508.20330, https://arxiv.org/abs/2508.20330 arXiv:2508.20330

work page arXiv 2025

[18] [18]

Neuroback: Improving CDCL SAT solving using graph neural networks

Wenxi Wang, Yang Hu, Mohit Tiwari, Sarfraz Khurshid, Kenneth McMillan, and Risto Miikkulainen. Neuroback: Improving CDCL SAT solving using graph neural networks. In The Twelfth International Conference on Learning Representations , 2024

work page 2024

[19] [19]

Understanding how dimension reduction tools work: an empirical approach to deciphering t-sne, umap, trimap, and pacmap for data visualization

Yingfan Wang, Haiyang Huang, Cynthia Rudin, and Yaron Shaposhnik. Understanding how dimension reduction tools work: an empirical approach to deciphering t-sne, umap, trimap, and pacmap for data visualization. Journal of Machine Learning Research , 22(201):1--73, 2021

work page 2021

[20] [20]

Vqgraph: Rethinking graph representation space for bridging gnns and mlps

Ling Yang, Ye Tian, Minkai Xu, Zhongyi Liu, Shenda Hong, Wei Qu, Wentao Zhang, Bin Cui, Muhan Zhang, and Jure Leskovec. Vqgraph: Rethinking graph representation space for bridging gnns and mlps. In International Conference on Learning Representations , 2024

work page 2024

[21] [21]

Nlocalsat: Boosting local search with solution prediction

Wenjie Zhang, Zeyu Sun, Qihao Zhu, Ge Li, Shaowei Cai, Yingfei Xiong, and Lu Zhang. Nlocalsat: Boosting local search with solution prediction. In Christian Bessiere, editor, Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI-20 , pages 1177--1183. International Joint Conferences on Artificial Intelligence Orga...

work page doi:10.24963/ijcai.2020/164 2020

[22] [22]

Grass: Combining graph neural networks with expert knowledge for sat solver selection

Zhanguang Zhang, Didier Ch \'e telat, Joseph Cotnareanu, Amur Ghose, Wenyi Xiao, Hui-Ling Zhen, Yingxue Zhang, Jianye Hao, Mark Coates, and Mingxuan Yuan. Grass: Combining graph neural networks with expert knowledge for sat solver selection. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages 6301--6311, 2024

work page 2024