Lifelong learning with deep neural networks is well-known to suffer from catastrophic forgetting: The performance on previous tasks drastically degrades when learning a new task. To alleviate this effect, we propose to leverage a large stream of unlabeled data easily obtainable in the wild. In particular, we design a novel class-incremental learning scheme with (a) a new distillation loss, termed global distillation, (b) a learning strategy to avoid overfitting to the most recent task, and (c) a confidence-based sampling method to effectively leverage unlabeled external data. Our experimental results on various datasets, including CIFAR and ImageNet, demonstrate the superiority of the proposed methods over prior methods, particularly when a stream of unlabeled data is accessible: Our method shows up to 15.8% higher accuracy and 46.5% less forgetting compared to the state-of-the-art method. The code is available at https://github.com/kibok90/iccv2019-inc.
|Title of host publication||Proceedings - 2019 International Conference on Computer Vision, ICCV 2019|
|Publisher||Institute of Electrical and Electronics Engineers Inc.|
|Number of pages||10|
|Publication status||Published - 2019 Oct|
|Event||17th IEEE/CVF International Conference on Computer Vision, ICCV 2019 - Seoul, Korea, Republic of|
Duration: 2019 Oct 27 → 2019 Nov 2
|Name||Proceedings of the IEEE International Conference on Computer Vision|
|Conference||17th IEEE/CVF International Conference on Computer Vision, ICCV 2019|
|Country/Territory||Korea, Republic of|
|Period||19/10/27 → 19/11/2|
Bibliographical noteFunding Information:
This work was supported in part by Kwanjeong Educational Foundation Scholarship, NSF CAREER IIS-1453651, and Sloan Research Fellowship. We also thank Lajanugen Logeswaran, Jongwook Choi, Yijie Guo, Wilka Carvalho, and Yunseok Jang for helpful discussions.
© 2019 IEEE.
All Science Journal Classification (ASJC) codes
- Computer Vision and Pattern Recognition