Unsupervised Learning for Combinatorial Optimization Needs Meta Learning

Authors: Haoyu Peter Wang, Pan Li

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the benefits of Meta-EGN via experiments within three benchmark CO problems (max clique, vertex cover, and max independent set) on multiple synthetic graphs and three real-world graph datasets, with the number of nodes ranging from 100 to 5000.
Researcher Affiliation Academia Haoyu Wang1, Pan Li1,2 1. Department of Electrical and Computer Engineering, Georgia Institute of Technology 2. Department of Computer Science, Purdue University
Pseudocode Yes Algorithm 1 Train Meta-EGN and Test Meta-EGN with/without Fine-tuning
Open Source Code Yes Our code is available at: https://github.com/Graph-COM/Meta_CO
Open Datasets Yes We conduct experiments on the MC and MVC problems over three real datasets Twitter (Leskovec & Krevl, 2014), COLLAB and IMDB (Yanardag & Vishwanathan, 2015) and two synthetic datasets with 200 and 500 nodes generated by the RB model (Xu, 2007).
Dataset Splits Yes For the real datasets, training/validation/test instances are randomly divided with the ratio of 7:1:2; For RB200 and RB500, 2000/100/100 graphs are generated for training/validation/test instances; ... We generate 15 graphs for validation and 20 graphs for each node degree configuration in D for test.
Hardware Specification Yes We run all experiments by using a Xeon(R) Gold 6248 CPU with 26 threads and a Quadro RTX 6000 GPU.
Software Dependencies Yes All the codes run on the Py Torch platform 1.9.0 (Paszke et al., 2019) and Py Torch Geometric framework 1.7.2 (Fey & Lenssen, 2019).
Experiment Setup Yes For the MC and MVC problems, we use 4-layer GIN (Xu et al., 2019) as the backbone network for both meta-EGN and EGN. We use 1e-3 as both the outer learning rate (γ) of Meta-EGN and the learning rate of EGN. ... For the MIS problem, we use 6-layer GIN. The outer learning rate (γ) of Meta-EGN and the learning rate of EGN are set as 1e-4. The inner learning rate (α) of Meta-EGN is always set as 5e-5.