Abstract

Modern machine learning suffers from \textit{catastrophic forgetting} when learning new classes incrementally. The performance dramatically degrades due to the missing data of old classes. Incremental learning methods have been proposed to retain the knowledge acquired from the old classes, by using knowledge distilling and keeping a few exemplars from the old classes. However, these methods struggle to \textbf{scale up to a large number of classes}. We believe this is because of the combination of two factors: (a) the data imbalance between the old and new classes, and (b) the increasing number of visually similar classes. Distinguishing between an increasing number of visually similar classes is particularly challenging, when the training data is unbalanced. We propose a simple and effective method to address this data imbalance issue. We found that the last fully connected layer has a strong bias towards the new classes, and this bias can be corrected by a linear model. With two bias parameters, our method performs remarkably well on two large datasets: ImageNet (1000 classes) and MS-Celeb-1M (10000 classes), outperforming the state-of-the-art algorithms by 11.1\% and 13.2\% respectively.

Keywords

ForgettingComputer scienceArtificial intelligenceScale (ratio)Machine learningSimple (philosophy)Class (philosophy)Training setIncremental learningAlgorithm

Affiliated Institutions

Related Publications

A Survey on Deep Transfer Learning

As a new classification platform, deep learning has recently received increasing attention from researchers and has been successfully applied to many domains. In some domains, l...

2018 Lecture notes in computer science 2764 citations

Publication Info

Year
2019
Type
article
Citations
1145
Access
Closed

External Links

Social Impact

Altmetric

Social media, news, blog, policy document mentions

Citation Metrics

1145
OpenAlex

Cite This

Yue Wu, Yinpeng Chen, Lijuan Wang et al. (2019). Large Scale Incremental Learning. . https://doi.org/10.1109/cvpr.2019.00046

Identifiers

DOI
10.1109/cvpr.2019.00046