AI Paper Atlas

Research map / Learning theory and algorithms / Combinatorial and search optimization

Unsupervised Learning for Combinatorial Optimization Needs Meta Learning

Haoyu Peter Wang, Pan Li

ICLR 2023 · Learning theory and algorithms › Combinatorial and search optimization · approach: Unsupervised neural combinatorial optimization · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Train with a meta-learning objective for instance-wise adaptation instead of minimizing average instance lossEGN · MC, TwitterApR: 0.926 ± 0.113(0.17) -> 0.976 ± 0.048(0.17)Improved
Train with a meta-learning objectiveEGN · MC, RB 200ApR: 0.820 ± 0.188 (0.26) -> 0.834 ± 0.178 (0.26)No clear effect
Train with a meta-learning objectiveEGN · MC, Twitter to RB2000ApR: 0.594±0.210(0.07) -> 0.690±0.201(0.07)Improved
Train with a meta-learning objectiveEGN · MVC, RB 500ApR: 1.021 ± 0.002 (0.48) -> 1.016 ± 0.002 (0.48)Improved
Add one-step fine-tuning at test time (ablation)Meta-EGN · MC, RB 500ApR: 0.834 ± 0.198 (0.29) -> 0.878 ± 0.181 (0.89)Improved
Add one-step fine-tuning at test time (ablation)Meta-EGN · MVC, TwitterApR: 1.019 ± 0.017(0.29) -> 1.017 ± 0.017(0.80)Improved
Use Meta-EGN to modify DGA solutionsDGA · MIS, RRGs, 10^5 nodesApR improvement: 0.0032; 0.0045; 0.0051; 0.0050Improved
Use Meta-EGN to modify RGA solutionsRGA · MIS, RRGs, 10^5 nodesApR improvement: 0.0871; 0.1045; 0.1083; 0.0969Improved
Reduce training data from 750 instances to 64 (ablation)Meta-EGN · MC, TwitterApR: 0.976±0.048 -> 0.970±0.058Hurt
Reduce training data from 750 instances to 64 (ablation)EGN · MC, TwitterApR: 0.926±0.113 -> 0.909±0.122Hurt
Train only on degree-3 RRGs rather than the mixed-degree training set (ablation)Meta-EGN · MIS, RRGs at other degreesApRHurt
Train only on degree-20 RRGs rather than the mixed-degree training set (ablation)Meta-EGN · MIS, RRGs at other degreesApRMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy