Graph Few-shot Class-incremental Learning
This addresses the incremental learning challenge for graph-based applications like social media and recommendation systems, but it is incremental as it builds on existing few-shot and class-incremental learning concepts.
The paper tackles the Graph Few-shot Class-incremental Learning problem, where a graph model must classify both new and old classes with limited data, and proposes a Graph Pseudo Incremental Learning paradigm and HAG-Meta framework, achieving remarkable advantages over baselines and state-of-the-art methods on three real-world datasets.
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. 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.