SMUG: Sand Mixing for Unobserved Class Detection in Graph Few-Shot Learning

Chenxu Wang, Xichan Nie, Jinfeng Chen, Pinghui Wang, Junzhou Zhao, Xiaohong Guan

Keywords: graph few-shot learning, out-of-distribution detection, sand mixing

TL;DR: This paper presents a novel graph few-shot learning framework with the ability of detecting unobserved classes.

Abstract: Graph few-shot learning (GFSL) has achieved great success in node classification tasks with sparse labels. However, graph few-shot classification (FSC) models often encounter the problem of classifying test samples with unobserved (and unknown) classes due to the rareness of labels. We formulate this problem as out-of-distribution (OOD) sample detection in inductive graph few-shot learning. This paper presents SMUG, a novel GFSL framework with the ability of detecting unobserved classes. Since we have no unobserved-class samples in a practical training dataset, it is challenging for the FSC model to retrieve the knowledge about unknown classes from labeled samples. To address this difficulty, we propose a sand mixing scheme to introduce an observed class as artificial OOD samples into meta-tasks. We also develop two unsupervised OOD discriminators to identify OOD samples. Thus, we can assess the performance of OOD discriminator since we know the true class of these artificial OOD samples. Subsequently, we design a novel training procedure to optimize the encoder based on the performance of the OOD discriminator and the FSC model. It not only enables the GFSL model to distinguish OOD samples but also promotes the classification accuracy of normal samples. We conduct extensive experiments to evaluate the effectiveness of SMUG based on four benchmark datasets. Experimental results demonstrate that SMUG achieves superior performance over state-of-the-art approaches in OOD detection and sample classification.

Track: Graph Algorithms and Learning for the Web

Submission Guidelines Scope: Yes

Submission Guidelines Blind: Yes

Submission Guidelines Format: Yes

Submission Guidelines Limit: Yes

Submission Guidelines Authorship: Yes

Student Author: No

Submission Number: 899