What question did this study set out to answer?

The aim is to enhance learning models' ability to adapt to new image classes with minimal data without forgetting previous classes.

May 16, 2026Open Access

VPAF-FSCIL: Virtual Prototype Calibration and Parameter-Adaptive Freezing for Few-Shot Class-Incremental Learning

Key Points

The aim is to enhance learning models' ability to adapt to new image classes with minimal data without forgetting previous classes.
Integrated a feature-decoupled network with a Parameter-Adaptive Freezing training strategy.
Decoupled the model into three components: Feature Extractor, Feature Distributor, and ETF Classifier.
Used the Fisher Information Matrix to selectively freeze parameters to minimize forgetting.
Achieved higher classification accuracy compared to state-of-the-art FSCIL methods.
Improved stability of feature representations during adaptation to new classes observed in extensive experiments.
Demonstrated effectiveness across multiple datasets including CIFAR-100, miniImageNet, and CUB-200-2011.

Abstract

Class Incremental Learning (CIL) enables models to learn new classes in images while retaining knowledge of previously learned classes. However, conventional CIL assumes abundant data for new classes, which is often unrealistic in real-world applications such as autonomous driving and medical diagnosis. Few-Shot Class Incremental Learning (FSCIL) addresses this issue by enabling the learning of new classes with minimal data. Nevertheless, it faces challenges like catastrophic forgetting, feature drift, and class imbalance, which can hinder effective adaptation. To tackle these challenges, we propose a method that integrates a feature-decoupled network architecture with a Parameter-Adaptive Freezing (PAF) training strategy, aiming to improve feature stability and adaptability to new classes. Inspired by Equiangular Tight Frame (ETF) theory, we decouple the model into three components: a Feature Extractor, a Feature Distributor, and an ETF Classifier. This structure integrates adaptive virtual class space allocation to ensure optimal embeddings for new classes. Moreover, PAF selectively freezes critical parameters based on the Fisher Information Matrix, which helps minimize forgetting of old classes while simultaneously improving generalization for new classes. Extensive experiments across CIFAR-100, miniImageNet, and CUB-200-2011 demonstrate that our proposed method outperforms state-of-the-art FSCIL methods in both classification accuracy and stability.

Read Full Paperexternally

Demander à l'IA

Bookmark

View Full Paper