arXiv：cs.LG· Pedro C. Vieira, Viacheslav Borovitskiy, Pedro Ribeiro·· 13 小时前
深度集成在图上真的能捕捉不确定性吗？研究发现 GNN 中的“认知坍缩”
Do Deep Ensembles Actually Capture Uncertainty in Graph Neural Networks?
AI 导读
一项针对消息传递图神经网络的研究发现，深度集成相较单模型仅有边际提升，且提升主要来自平均化点预测中的优化噪声，而非更优的不确定性估计。研究在八个数据集上通过 aleatoric-epistemic 分解识别出“认知坍缩”：独立训练的网络预测高度趋同，使集成赖以捕捉认知不确定性的分歧机制失效；作者进一步推导出认知方差与图拓扑和 Dirichlet 能量预算的界，将其归因于消息传递的平滑偏差。
正文
Abstract:While deep ensembles are widely considered to be the default method for uncertainty quantification in deep learning, their effectiveness for graph-structured data is often simply assumed based on successes in domains like computer vision. We investigate standard deep ensembles specifically for message-passing graph neural networks. Benchmarking across eight datasets representing varied tasks and complexities, we reveal that ensembles provide only marginal improvement over a single model. These marginal gains stem primarily from averaging out optimization noise in point predictions rather than from meaningfully better uncertainty estimates. Through an aleatoric-epistemic decomposition, we identify epistemic collapse: independently trained networks consistently converge to overly similar predictions. Because disagreement is the fundamental mechanism through which ensembles capture epistemic uncertainty, this lack of diversity neutralizes their key advantage. Towards explaining this phenomenon, we derive a bound on epistemic variance in terms of the graph topology and Dirichlet energy budget, linking epistemic collapse to the smoothing bias of message passing. Our results suggest that deep ensemble success does not seamlessly transfer to graph machine learning.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2605.22593 [cs.LG] |
| (or arXiv:2605.22593v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2605.22593 arXiv-issued DOI via DataCite |
Submission history
From: Pedro C. Vieira [view email]
[v1]
Thu, 21 May 2026 15:07:47 UTC (1,555 KB)
[v2]
Fri, 9 Oct 2026 09:33:24 UTC (1,734 KB)
来源：arXiv：cs.LG · arxiv.org