Similarity-Navigated Conformal Prediction for Graph Neural Networks
Jianqing Song, Jianguo Huang, Wenyu Jiang, Baoming Zhang, Shuangjie Li, Chongjun Wang
TL;DR
This work tackles the lack of reliable uncertainty estimates for GNN-based node classification by applying conformal prediction (CP) with a novel similarity-guided score aggregation. The authors introduce SNAPS, which adaptively aggregates non-conformity scores from nodes likely to share the same label—guided by feature similarity and one-hop structure—to produce compact, valid CP prediction sets with higher singleton hit ratios. They provide exchangeability-based theoretical guarantees and demonstrate empirically that SNAPS reduces average prediction-set size while preserving coverage across ten graph datasets and even ImageNet, outperforming APS, DAPS, and CF-GNN in efficiency. The method extends naturally to image classification and offers a practical, scalable post-processing tool for reliable uncertainty quantification in graph-structured and related data. Limitations include the transductive setting and the computational cost of selecting similar-label nodes, pointing to future work on inductive extensions and more efficient similarity mechanisms.
Abstract
Graph Neural Networks have achieved remarkable accuracy in semi-supervised node classification tasks. However, these results lack reliable uncertainty estimates. Conformal prediction methods provide a theoretical guarantee for node classification tasks, ensuring that the conformal prediction set contains the ground-truth label with a desired probability (e.g., 95%). In this paper, we empirically show that for each node, aggregating the non-conformity scores of nodes with the same label can improve the efficiency of conformal prediction sets while maintaining valid marginal coverage. This observation motivates us to propose a novel algorithm named Similarity-Navigated Adaptive Prediction Sets (SNAPS), which aggregates the non-conformity scores based on feature similarity and structural neighborhood. The key idea behind SNAPS is that nodes with high feature similarity or direct connections tend to have the same label. By incorporating adaptive similar nodes information, SNAPS can generate compact prediction sets and increase the singleton hit ratio (correct prediction sets of size one). Moreover, we theoretically provide a finite-sample coverage guarantee of SNAPS. Extensive experiments demonstrate the superiority of SNAPS, improving the efficiency of prediction sets and singleton hit ratio while maintaining valid coverage.