Embedding Space Allocation with Angle-Norm Joint Classifiers for Few-Shot Class-Incremental Learning

2024-11-14

Neural Networks

AI Summary

Few-shot class-incremental learning (FSCIL) suffers from catastrophic forgetting caused by feature space interference and insufficient training on few-shot classes. This paper proposes a class-center-guided subspace allocation framework with an angle-norm joint classifier. First, we introduce a novel angle-norm joint logits mechanism that decouples discriminative feature learning from confidence estimation. Second, we dynamically partition the embedding space into class-specific subspaces anchored to predefined class centers, enabling feature disentanglement and imbalance-aware representation learning. Third, we design a norm-adaptive decision strategy that is compatible with nearest-class-mean (NCM) inference and supports lightweight plug-and-play deployment. Evaluated on standard benchmarks—including CIFAR-100 and ImageNet-Subset—our method achieves state-of-the-art performance, significantly improving both accuracy and stability over prevailing FSCIL approaches.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Representation Learning for Vision
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization

Abstract

Few-shot class-incremental learning (FSCIL) aims to continually learn new classes from only a few samples without forgetting previous ones, requiring intelligent agents to adapt to dynamic environments. FSCIL combines the characteristics and challenges of class-incremental learning and few-shot learning: (i) Current classes occupy the entire feature space, which is detrimental to learning new classes. (ii) The small number of samples in incremental rounds is insufficient for fully training. In existing mainstream virtual class methods, to address the challenge (i), they attempt to use virtual classes as placeholders. However, new classes may not necessarily align with the virtual classes. For challenge (ii), they replace trainable fully connected layers with Nearest Class Mean (NCM) classifiers based on cosine similarity, but NCM classifiers do not account for sample imbalance issues. To address these issues in previous methods, we propose the class-center guided embedding Space Allocation with Angle-Norm joint classifiers (SAAN) learning framework, which provides balanced space for all classes and leverages norm differences caused by sample imbalance to enhance classification criteria. Specifically, for challenge (i), SAAN divides the feature space into multiple subspaces and allocates a dedicated subspace for each session by guiding samples with the pre-set category centers. For challenge (ii), SAAN establishes a norm distribution for each class and generates angle-norm joint logits. Experiments demonstrate that SAAN can achieve state-of-the-art performance and it can be directly embedded into other SOTA methods as a plug-in, further enhancing their performance.

Problem

• Allocate embedding space for few-shot incremental learning
• Address sample imbalance in incremental class learning
• Enhance classification with angle-norm joint classifiers

Innovation

• Divides feature space into dedicated subspaces
• Uses angle-norm joint logits for classification
• Guides samples with pre-set category centers