Abstract

两个问题，1）灾难性遗忘；2）As new observations of old classes come sequentially over time, the distribution may change in unforeseen way, making the performance degrade dramatically on future data, which is referred to as concept drift.
一个新的online learning scenario and handle both new classes data and new observations of old classes

Introduction

process data sequentially in an online Learning mode《Incremental online learning: A review and comparison of state-of-the-art algorithms, Neurocomputing 2018》;
The second issue, concept drift《A survey on concept drift adaptation, ACM computation Surveys 2014》;the data of already learned classes may change in unforeseen ways《Classifier adaptation at prediction time, CVPR2015》;

Related Work

Conclusion

提出一个modified cross-distillation loss；更新exemplar set的方法
quality一般；一个更实际online food image classification based on our complete framework using the Food-101 dataset;方法设计incremental；insight比较简单；没有开源code