Abstract:Few-shot class-incremental learning (FSCIL) aims to build machine learning model that can continually learn new concepts from a few data samples, without forgetting knowledge of old classes. The challenges of FSCIL lies in the limited data of new classes, which not only lead to significant overfitting issues but also exacerbates the notorious catastrophic forgetting problems. As proved in early studies, building sample relationships is beneficial for learning from few-shot samples. In this paper, we promote the idea to the incremental scenario, and propose a Sample-to-Class (S2C) graph learning method for FSCIL. Specifically, we propose a Sample-level Graph Network (SGN) that focuses on analyzing sample relationships within a single session. This network helps aggregate similar samples, ultimately leading to the extraction of more refined class-level features. Then, we present a Class-level Graph Network (CGN) that establishes connections across class-level features of both new and old classes. This network plays a crucial role in linking the knowledge between different sessions and helps improve overall learning in the FSCIL scenario. Moreover, we design a multi-stage strategy for training S2C model, which mitigates the training challenges posed by limited data in the incremental process. The multi-stage training strategy is designed to build S2C graph from base to few-shot stages, and improve the capacity via an extra pseudo-incremental stage. Experiments on three popular benchmark datasets show that our method clearly outperforms the baselines and sets new state-of-the-art results in FSCIL.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to continuously learn new concepts in Few - Shot Class - Incremental Learning (FSCIL) with only a small number of new - class samples, while not forgetting the knowledge of old classes. There are mainly two challenges in FSCIL: 1. **Limited new - class data**: This will not only lead to significant over - fitting problems, but also exacerbate the problem of catastrophic forgetting, that is, the model forgets the previously learned knowledge when learning new tasks. 2. **Importance of constructing sample relationships**: Early studies have proved that constructing relationships between samples is beneficial for learning from a small number of samples. However, in the incremental learning scenario, this method of constructing sample relationships has not been fully explored. To address these challenges, the paper proposes a new Sample - to - Class (S2C) graph learning method, specifically including: - **Sample - level Graph Network (SGN)**: Focus on analyzing the sample relationships within a single session, helping to aggregate similar samples and extract more refined class features. - **Class - level Graph Network (CGN)**: Establish connections between new and old class features, link the knowledge between different sessions, and improve the overall learning effect. - **Multi - stage training strategy**: A training strategy with three main stages is designed to address the training challenges brought by limited data in the incremental process: - **Basic graph pre - construction stage**: Initialize the CGN with a large amount of data in the basic session to retain a large amount of prior knowledge. - **S2C pseudo - incremental training stage**: Generate virtual FSCIL tasks by randomly sampling FSL tasks on the basic data set to adapt to FSL tasks. - **Few - shot incremental training stage**: Apply the S2C model to real FSCIL tasks to further optimize the model performance. Experimental results show that this method significantly outperforms the baseline methods on three popular benchmark data sets (CIFAR100, miniImageNet, CUB200) and achieves new state - of - the - art results.

Constructing Sample-to-Class Graph for Few-Shot Class-Incremental Learning

Two-level Graph Network for Few-Shot Class-Incremental Learning

Few-Shot Incremental Learning with Continually Evolved Classifiers

A Cognition-Driven Framework for Few-Shot Class-Incremental Learning

Grassmann Graph Embedding for Few-Shot Class Incremental Learning.

A survey on few-shot class-incremental learning

Dynamic Support Network for Few-Shot Class Incremental Learning

Jointly Optimized Classifiers for Few-Shot Class-Incremental Learning

Compositional Few-Shot Class-Incremental Learning

Inductive Graph Few-shot Class Incremental Learning

Few-shot class-incremental learning based on representation enhancement

Pseudo Initialization Based Few-Shot Class Incremental Learning

Analogical Learning-Based Few-Shot Class-Incremental Learning

S3C: Self-Supervised Stochastic Classifiers for Few-Shot Class-Incremental Learning

Enhanced Few-Shot Class-Incremental Learning via Ensemble Models

Self-supervised Contrastive Feature Refinement for Few-Shot Class-Incremental Learning.

Few-Shot Class-Incremental Learning Via Class-Aware Bilateral Distillation

MetaFSCIL: A Meta-Learning Approach for Few-Shot Class Incremental Learning

Knowledge Representation by Generic Models for Few-Shot Class-Incremental Learning

Rethinking Few-shot Class-incremental Learning: Learning from Yourself

Knowledge Adaptation Network for Few-Shot Class-Incremental Learning