Technology Category
Machine Learning: Multi-class/Multi-label Learning & Extreme ClassificationSearch and Optimization: Learning to SearchComputer Vision: Learning & Optimization for CV
Controllable Forgetting Mechanism for Few-Shot Class-Incremental Learning
🤖 AI Summary
This paper addresses the forgetting–adaptation trade-off in ultra-low-shot (e.g., 1-shot) class-incremental learning, where rapid adaptation to novel classes must be reconciled with strong retention of previously learned knowledge. We propose a novel class detection (NCD) rule that enables *a priori*, controllable regulation of forgetting—a first in this setting—and introduce new evaluation metrics, NCR@2FOR and NCR@5FOR, to quantitatively assess balanced performance. Our method operates via discriminative modeling in feature space and employs a lightweight, parameter-free rule-based mechanism, requiring no auxiliary networks or replay memory, and is fully compatible with mainstream few-shot class-incremental learning (FSCIL) frameworks. On CIFAR-100 under the 1-shot single-new-class setting, our approach achieves up to a 30% improvement in novel-class accuracy while strictly limiting average forgetting on base classes to under 2%, significantly outperforming existing methods.
Technology Category
Application Category
📝 Abstract
Class-incremental learning in the context of limited personal labeled samples (few-shot) is critical for numerous real-world applications, such as smart home devices. A key challenge in these scenarios is balancing the trade-off between adapting to new, personalized classes and maintaining the performance of the model on the original, base classes. Fine-tuning the model on novel classes often leads to the phenomenon of catastrophic forgetting, where the accuracy of base classes declines unpredictably and significantly. In this paper, we propose a simple yet effective mechanism to address this challenge by controlling the trade-off between novel and base class accuracy. We specifically target the ultra-low-shot scenario, where only a single example is available per novel class. Our approach introduces a Novel Class Detection (NCD) rule, which adjusts the degree of forgetting a priori while simultaneously enhancing performance on novel classes. We demonstrate the versatility of our solution by applying it to state-of-the-art Few-Shot Class-Incremental Learning (FSCIL) methods, showing consistent improvements across different settings. To better quantify the trade-off between novel and base class performance, we introduce new metrics: NCR@2FOR and NCR@5FOR. Our approach achieves up to a 30% improvement in novel class accuracy on the CIFAR100 dataset (1-shot, 1 novel class) while maintaining a controlled base class forgetting rate of 2%.
Problem
Research questions and friction points this paper is trying to address.
Few-shot Learning
Knowledge Balance
Incremental Learning
Innovation
Methods, ideas, or system contributions that make the work stand out.
Incremental Learning
Knowledge Retention
Minimal Forgetting
🔎 Similar Papers
No similar papers found.
