Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

FOSTER: Feature Boosting and Compression for Class-Incremental Learning

About

The ability to learn new concepts continually is necessary in this ever-changing world. However, deep neural networks suffer from catastrophic forgetting when learning new categories. Many works have been proposed to alleviate this phenomenon, whereas most of them either fall into the stability-plasticity dilemma or take too much computation or storage overhead. Inspired by the gradient boosting algorithm to gradually fit the residuals between the target model and the previous ensemble model, we propose a novel two-stage learning paradigm FOSTER, empowering the model to learn new categories adaptively. Specifically, we first dynamically expand new modules to fit the residuals between the target and the output of the original model. Next, we remove redundant parameters and feature dimensions through an effective distillation strategy to maintain the single backbone model. We validate our method FOSTER on CIFAR-100 and ImageNet-100/1000 under different settings. Experimental results show that our method achieves state-of-the-art performance. Code is available at: https://github.com/G-U-N/ECCV22-FOSTER.

Fu-Yun Wang, Da-Wei Zhou, Han-Jia Ye, De-Chuan Zhan• 2022

Related benchmarks

TaskDatasetResultRank
Class-incremental learningCIFAR-100
Averaged Incremental Accuracy78.07
234
Class-incremental learningImageNet-R
Average Accuracy76.8
103
Class-incremental learningCIFAR-100 10 (test)
Average Top-1 Accuracy72.2
75
Class-incremental learningImageNet-100
Avg Acc82.79
74
Class-incremental learningCIFAR100 B50 (test)
Average Accuracy67.94
67
Class-incremental learningCIFAR100-LT rho=100 (test)
Avg Acc37.91
48
Class-incremental learningImageNet Subset-LT rho=100 (test)
Accuracy46.51
48
Class-incremental learningFood101-LT rho=100 (test)
Accuracy32.39
48
Class-incremental learningCUB
Avg Accuracy86.9
45
Class-incremental learningImageNet-100 B=50, C=10 1.0
Avg Incremental Acc84.54
42
Showing 10 of 38 rows

Other info

Code

Follow for update