Research map / Graph learning / Graph algorithms and clustering
Neural combinatorial optimization on graphs
Guide combinatorial and integer-programming solvers with graph neural networks.
159 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Graph algorithms and clustering, keyword cluster “search · directed · causal”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
42 papers in 2023–24 and 48 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 21.4% to 23.9%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Active Learning for Delineation of Curvilinear Structures (CVPR 2016 · 12 citations)
- 2017 · Learning Combinatorial Optimization Algorithms over Graphs (NeurIPS 2017 · 1,066 citations)
- 2018 · Learning Steady-States of Iterative Algorithms over Graphs (ICML 2018 · 110 citations)
- 2019 · Exact Combinatorial Optimization with Graph Convolutional Neural Networks (NeurIPS 2019 · 249 citations)
- 2020 · Accelerating Primal Solution Findings for Mixed Integer Programs Based on Solution Prediction (AAAI 2020 · 72 citations)
- 2021 · Inferring CAD Modeling Sequences Using Zone Graphs (CVPR 2021 · 50 citations)
- 2022 · MIP-GNN: A Data-Driven Framework for Guiding Combinatorial Solvers (AAAI 2022 · 34 citations)
- 2023 · AutoGraph: Optimizing DNN Computation Graph for Parallel GPU Kernel Execution (AAAI 2023 · 12 citations)
- 2024 · Learning to Optimize Permutation Flow Shop Scheduling via Graph-Based Imitation Learning (AAAI 2024 · 11 citations)
- 2025 · Destroy and Repair Using Hyper-Graphs for Routing (AAAI 2025 · 3 citations)
- 2026 · Learn from Global Correlations: Enhancing Evolutionary Algorithm via Spectral GNN (AAAI 2026 · 7 citations)
Earliest papers here
- Active Learning for Delineation of Curvilinear Structures (CVPR 2016 · 12 citations)
- Learning Combinatorial Optimization Algorithms over Graphs (NeurIPS 2017 · 1,066 citations)
- Deep Biaffine Attention for Neural Dependency Parsing (ICLR 2017 · 702 citations)
What has been tried
111 papers here report 1,068 measured changes (a component added, removed or swapped, compared with a baseline): helped 68%, no clear effect 9%, hurt 19%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Replace the LSTM encoder with GGCN | PFSS-20 · makespan | 31.0 -> 30.2 | Helped | Learning to Optimize Permutation Flow Shop Scheduling via Graph-Based Imitation Learning (AAAI 2024 · 11 citations) |
| Add CAM to TPM | JSSP 10×10 training · Convergence | Helped | Deep Reinforcement Learning Guided Improvement Heuristic for Job Shop Scheduling (ICLR 2024 · 11 citations) | |
| add output shape features | Inception-V3 · speedup (%) | 8.59 -> 17.9 | Helped | A Structure-Aware Framework for Learning Device Placements on Computation Graphs (NeurIPS 2024 · 7 citations) |
| apply graph-based smoothing to GGS | GFP hard · Median Fitness | 0.10 -> 0.74 | Helped | Improving protein optimization with smoothed fitness landscapes (ICLR 2024 · 4 citations) |
| use training sample size [20, 80] instead of [30, 70] | TSP200, TSP500, TSP1K, TSP5K, and TSP10K · Gap | 0.014% -> 0.0098%; 0.127% -> 0.113%; 0.268% -> 0.258%; 1.47% -> 1.42%; 3.06% -> 2.85% | Helped | Destroy and Repair Using Hyper-Graphs for Routing (AAAI 2025 · 3 citations) |
| add knowledge distillation, replacing the teacher GCN with a student GCN | synthetic and real-world datasets · inference time | Helped | COMBHelper: A Neural Approach to Reduce Search Space for Graph Combinatorial Problems (AAAI 2024 · 3 citations) | |
| Add path encoding to GRAF | CIFAR-10 on NB101, NB201, and NB301; CIFAR-100 on NB201; largest sample size · Kendall tau | Mixed | Surprisingly Strong Performance Prediction with Neural Graph Features (ICML 2024 · 2 citations) | |
| Add a secondary scoring model to select among beam-search candidate routes | Grid world · semantic routing performance | Helped | Semantic Routing via Autoregressive Modeling (NeurIPS 2024 · 2 citations) | |
| Use only tight cuts for training labels | Facility Location medium · IGC ratio (reduced/full) | 1.54 -> 1.76 | Helped | Learn2Aggregate: Supervised Generation of Chvatal-Gomory Cuts Using Graph Neural Networks (AAAI 2025 · 1 citation) |
| Add graph state across expansion iterations (GATState vs GAT) | Generated SR test problems, k = 2 · success rate | 9.04 -> 10.61 | Helped | Neuro-Guided Graph Search for Symbolic Regression (Student Abstract) (AAAI 2025 · 1 citation) |
| Remove the equidistribution regularizer | Poisson Square · Error reduction (%) | 23.99 -> 22.42 | Hurt | G-Adaptivity: optimised graph-based mesh relocation for finite element methods (ICML 2025 · 1 citation) |
| Remove priority regularization | ER4, 30 nodes · AUC | 0.984 ± 0.02 -> 0.937 ± 0.05 | Hurt | Constraint-Free Structure Learning with Smooth Acyclic Orientations (ICLR 2024 · 1 citation) |
What reviewers will expect
Datasets and baselines named in the 22 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Baselines
- POMO · 2 of 22 papers
- MVMoE · 2 of 22 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 90 papers; from OpenAlex and the papers' affiliations).
- Chinese Academy of Sciences · 4 papers
- Huawei · 4 papers
- National University of Singapore · 3 papers
- Alibaba Group · 3 papers
- Qualcomm · 3 papers
- Amazon · 3 papers
- Eindhoven University of Technology · 3 papers
- University of Cambridge · 3 papers
- Google · 3 papers
- Shanghai Jiao Tong University · 3 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Graph comparison and random features (155 papers)
- Graph embedding and spectral sparsification (126 papers)
- Continuous DAG structure learning (102 papers)
- Graph distance and retrieval (79 papers)
Elsewhere in the topic
- Neural and local graph clustering (81 papers)
- Deep learning for graph matching (42 papers)
- Optimization algorithms for graph matching (42 papers)
- Anchor-based multi-view clustering (31 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).