Graph Few-shot Class-incremental Learning

By Zhen Tan, Kaize Ding, Ruocheng Guo, Huan Liu. 2021

74 citations recorded by the source corpus.

The ability to incrementally learn new classes is vital to all real-world artificial intelligence systems. A large portion of high-impact applications like social media, recommendation systems, E-commerce platforms, etc. can be represented by graph models. In this paper, we investigate the challenging yet practical problem,Graph Few-shot Class-incremental (Graph FCL) problem, where the graph model is tasked to classify both newly encountered classes and previously learned classes. Towards that purpose, we put forward a Graph Pseudo Incremental Learning paradigm by sampling tasks recurrently from the base classes, so as to produce an arbitrary number of training episodes for our model to practice the incremental learning skill. Furthermore, we design a Hierarchical-Attention-based Graph Meta-learning framework, HAG-Meta from an optimization perspective. We present a task-sensitive regularizer calculated from task-level attention and node class prototypes to mitigate overfitting onto either novel or base classes. To employ the topological knowledge, we add a node-level attention module to adjust the prototype representation. Our model not only achieves greater stability of old knowledge consolidation, but also acquires advantageous adaptability to new knowledge with very limited data samples. Extensive experiments on three real-world datasets, including Amazon-clothing, Reddit, and DBLP, show that our framework demonstrates remarkable advantages in comparison with the baseline and other related state-of-the-art methods.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Jaya Kalla
2. Xin Ning
3. Weijun Li
4. Dacheng Tao
5. Prayag Tiwari
6. Soma Biswas
7. Hang Ran
8. Dongjin Song

In Computer Vision and Pattern Recognition

1. Jaya Kalla
2. Xin Ning
3. Weijun Li
4. Dacheng Tao
5. Prayag Tiwari
6. Soma Biswas
7. Hang Ran
8. Dongjin Song

Citations in the corpus

• S3C: Self-Supervised Stochastic Classifiers for Few-Shot Class-Incremental Learning (2023, methodology): “The works in [11,34,44] used graph-based networks for old classes’ knowledge retention.”
• CGLB: Benchmark Tasks for Continual Graph Learning (2022, background): “Finally, the essential difference between CGL, dynamic graph learning [14, 48, 18, 60, 35, 66, 33, 13, 5], and few-shot graph learning [65, 16, 57, 43] is worth noting.”
• A Survey on Few-Shot Class-Incremental Learning (2023, methodology): “Recent studies have applied FSCIL to graphs (Lu et al., 2022; Tan et al., 2022).”