[논문 리뷰] Adaptive Graph Convolutional Neural Networks

Ruoyu Li, Sheng Wang외 2명|arXiv (Cornell University)|2018. 01. 10.

Advanced Graph Neural Networks참고 문헌 23인용 수 95

한 줄 요약

AGCN은 잔여 그래프 라플라시안과 거리 메트릭을 학습하여 작업별 적응 그래프를 학습하고, 임의의 그래프 구조에서 컨볼루션을 가능하게 하여 성능과 수렴 속도를 개선합니다.

ABSTRACT

Graph Convolutional Neural Networks (Graph CNNs) are generalizations of classical CNNs to handle graph data such as molecular data, point could and social networks. Current filters in graph CNNs are built for fixed and shared graph structure. However, for most real data, the graph structures varies in both size and connectivity. The paper proposes a generalized and flexible graph CNN taking data of arbitrary graph structure as input. In that way a task-driven adaptive graph is learned for each graph data while training. To efficiently learn the graph, a distance metric learning is proposed. Extensive experiments on nine graph-structured datasets have demonstrated the superior performance improvement on both convergence speed and predictive accuracy.

연구 동기 및 목표

• 고정된 그래프가 아닌 다양한 그래프 구조를 처리하도록 그래프 CNN을 고무한다.
• 적응적이고 샘플-특이적인 그래프 라플라시안을 학습하는 스펙트럴 그래프 컨볼루션 계층을 제안한다.
• 적응 그래프를 구성하기 위한 거리-메트릭 학습을 도입한다.
• 학습 효율성을 향상시키기 위해 특징 공간 재매개변화 및 잔여 그래프 구성 요소를 도입한다.
• 분자 및 포인트 클라우드를 포함한 다중 그래프 구조 데이터셋에서 우수한 성능을 입증한다.]
• method:[
• Introduce SGC-LL: a spectral graph convolution layer with an adaptive Laplacian learned via a distance metric.
• Use Mahalanobis-like distance D(xi,xj) = sqrt((xi-xj)^T M (xi-xj)) with M = Wd Wd^T as trainable.
• Compute a residual graph Laplacian update L_res from learned metric and features, and form \tilde L = L + alpha L_res.
• Represent spectral filter g_theta(L) as a polynomial in updated Laplacian using Chebyshev expansion for efficiency.
• Apply a feature-space transform Y = U g_theta(L) U^T X with trainable W and b to embed intra- and inter-vertex features.
• Adopt per-layer residual graph updates to allow batch training over graphs with differing topology and size.]
• research_questions:[
• Can a graph CNN handle data with arbitrary graph structure and size without losing information due to fixed graphs?
• Does learning a per-sample adaptive graph Laplacian improve predictive accuracy and convergence speed?
• Can distance-metric learning effectively customize graph topology for a given task?
• Does re-parameterization of vertex features enhance graph convolution performance?
• How does the proposed AGCN perform on molecular, point-cloud, and multi-task toxicity datasets compared to existing graph CNNs?]
• key_findings:[
• AGCN outperforms state-of-the-art graph CNNs on multiple datasets, showing faster convergence and better predictive accuracy.
• Learning a residual Laplacian L_res per sample yields graph topologies that better serve the prediction task and can introduce new edges beyond intrinsic graphs.
• Distance-metric learning with a low parameter cost (O(d^2) or O(d)) makes topology updates computationally efficient and graph-size independent.
• The model supports training on batches of diverse graphs with different sizes, thanks to shared feature transforms and metrics.
• Experiments on Delaney, Az-logD, NIH-NCI, Hydration-free energy, Tox21, ClinTox, Sider, Toxcast, and Sydney point clouds demonstrate broad effectiveness of AGCN across regression, classification, and multi-task scenarios.
• Table 1 shows AGCN achieving lower RMSE than graphconv, NFP, and GCN across several molecular datasets.

실험 결과

주요 결과

데이터 세트	그래프 컨볼루션 RMSE	NFP RMSE	GCN RMSE	AGCN RMSE
Delaney	0.4222 ± 8.38e-2	0.4955 ± 2.30e-3	0.4665 ± 2.07e-3	0.3061 ± 5.34e-3
Az-logD	0.7516 ± 8.42e-3	0.9597 ± 5.70e-3	1.0459 ± 3.92e-3	0.7362 ± 3.54e-3
NCI	0.8695 ± 3.55e-3	0.8748 ± 7.50e-3	0.8717 ± 4.14e-3	0.8647 ± 4.67e-3
Hydration-free energy	2.0329 ± 2.70e-2	3.4082 ± 3.95e-2	2.2868 ± 1.37e-2	1.3317 ± 2.73e-2