AutoCoG: A Unified Data-Model Co-Search Framework for Graph Neural Networks

Duc N.M Hoang, Kaixiong Zhou, Tianlong Chen, Xia Hu, Zhangyang Wang

Abstract: Neural architecture search (NAS) has demonstrated success in discovering promising architectures for vision or language modeling tasks, and it has recently been introduced to searching for graph neural networks (GNNs) as well. Despite the preliminary success, GNNs struggle in dealing with heterophily or low-homophily graphs where connected nodes may have different class labels and dissimilar features. To this end, we propose co-optimizing both the input graph topology and the model's architecture topology simultaneously. That yields AutoCoG, the first unified data-model co-search NAS framework for GNNs. By defining a highly flexible data-model co-search space, AutoCoG is gracefully formulated as a principled bi-level optimization that can be end-to-end solved by the differentiable search methods. Experiments show AutoCoG achieves gains of up to 4% for Actor, 7.3% on average for Web datasets, 0.17% for CoAuthor-CS, and finally 5.4% for Wikipedia-Photo benchmarks. All codes will be released upon paper acceptance.

Keywords: GCN, NAS

One-sentence Summary: We propose AutoCoG, a NAS framework to unified model's architecture search and graph-augmentation in a differentiable manner.

Track: Main track

Reproducibility Checklist: Yes

Broader Impact Statement: Yes

Paper Availability And License: Yes

Code Of Conduct: Yes

Reviewers: duc hoang, [e-mail]

CPU Hours: 0

GPU Hours: 48

TPU Hours: 0

Evaluation Metrics: No

Class Of Approaches: Gradient-based Methods, Progressive Search

Datasets And Benchmarks: WebKB-Texas, WebKB-Wisconsin, WebKB-Cornell, CoAuthor-CS, CoAuthor-Physics, Amazon-Photos, Amazon-Computers, Actor

Performance Metrics: Accuracy

Estimated CO2e Footprint: 5.18

Benchmark Performance:

Actor 38.04
Texas 80.27
Wisconsin 80.39
Cornell 64.86
Computer 78.91
CS 92.05
Photos 85.16
Physics 93.28