AI Paper Atlas

Research map / Graph learning / Graph algorithms and clustering

Learning to Solve NP-Complete Problems: A Graph Neural Network for Decision TSP

Marcelo Prates, Pedro H. C. Avelar, Henrique Lemos, Luis C. Lamb, Moshe Y. Vardi

AAAI 2019 · Graph learning › Graph algorithms and clustering · approach: Neural combinatorial optimization on graphs · 180 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 paired decision examples at ±2% of the optimumGNN-TSP · Decision TSP, test (n=20–40)accuracy: 80%Improved
Increase the decision-cost deviation from 2% to 5% or 10%GNN-TSP · Decision TSP, test (n=20–40)accuracy: 80 -> 98; 80 -> 100Improved
Reduce the decision-cost deviation from 2% to 1%GNN-TSP · Decision TSP, test (n=20–40)accuracy: 80 -> 66Hurt
Apply the model to random metric rather than Euclidean graphsGNN-TSP · Decision TSP, 2% deviationaccuracy: 80 -> 64Hurt
Apply the model to non-metric random graphsGNN-TSP · Decision TSP, random graphsaccuracy: 50Hurt
Increase instance size beyond the training rangeGNN-TSP · Decision TSP, larger instancesaccuracy: 50%Hurt
Increase deviation for larger instancesGNN-TSP · Decision TSP, larger instancesaccuracy: >95%Improved
Add ±100%, +200%, and +1000% examples in a one-epoch retraining passGNN-TSP · Decision TSP, test accuracyaccuracyNo clear effect
Use binary search over target costs with the trained decision modelGNN-TSP · Route cost prediction, test (n=20–40)absolute deviation: 1.5%Improved
Use the trained model rather than Nearest Neighbor or Simulated Annealing as a decision predictorGNN-TSP · Decision TSP, tested deviationstrue positive rate: >90% TPR from deviations 4% and aboveImproved
Use GNN-TSP rather than Simulated Annealing for TSPLIB route-cost estimationGNN-TSP with binary search · TSPLIB instancesabsolute relative deviation: 6 out of 9Mixed

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