Abstract:Regularization-based methods are beneficial to alleviate the catastrophic forgetting problem in class-incremental learning. With the absence of old task images, they often assume that old knowledge is well preserved if the classifier produces similar output on new images. In this paper, we find that their effectiveness largely depends on the nature of old classes: they work well on classes that are easily distinguishable between each other but may fail on more fine-grained ones, e.g., boy and girl. In spirit, such methods project new data onto the feature space spanned by the weight vectors in the fully connected layer, corresponding to old classes. The resulting projections would be similar on fine-grained old classes, and as a consequence the new classifier will gradually lose the discriminative ability on these classes. To address this issue, we propose a memory-free generative replay strategy to preserve the fine-grained old classes characteristics by generating representative old images directly from the old classifier and combined with new data for new classifier training. To solve the homogenization problem of the generated samples, we also propose a diversity loss that maximizes Kullback Leibler (KL) divergence between generated samples. Our method is best complemented by prior regularization-based methods proved to be effective for easily distinguishable old classes. We validate the above design and insights on CUB-200-2011, Caltech-101, CIFAR-100 and Tiny ImageNet and show that our strategy outperforms existing memory-free methods with a clear margin. Code is available at <a class="link-external link-https" href="https://github.com/xmengxin/MFGR" rel="external noopener nofollow">this https URL</a>

Incremental Classifier Learning with Generative Adversarial Networks

Incremental Learning Using Conditional Adversarial Networks

Class Incremental Learning Via Multi-hinge Distillation

Memory-Free Generative Replay For Class-Incremental Learning

Incremental Concept Learning via Online Generative Memory Recall

Incremental Focal Loss GANs.

A Strategy for an Uncompromising Incremental Learner

Incremental learning for the detection and classification of GAN-generated images

Large Scale Incremental Learning

Bypassing Logits Bias in Online Class-Incremental Learning with a Generative Framework

Learning a Dual-Branch Classifier for Class Incremental Learning

More Classifiers, Less Forgetting: A Generic Multi-classifier Paradigm for Incremental Learning

On the Exploration of Incremental Learning for Fine-grained Image Retrieval

Incremental Learning of Object Detectors without Catastrophic Forgetting

Generative Feature Replay For Class-Incremental Learning

Class-Incremental Learning Based on Big Dataset Pre-Trained Models

Class incremental learning with probability dampening and cascaded gated classifier

Overcoming Catastrophic Forgetting in Federated Class-Incremental Learning via Federated Global Twin Generator

Maintaining Discrimination and Fairness in Class Incremental Learning

DCIGAN - A Distributed Class-Incremental Learning Method Based on Generative Adversarial Networks.

DeepCollaboration: Collaborative Generative and Discriminative Models for Class Incremental Learning