Self-Supervised Learning Aided Class-Incremental Lifelong Learning
By Song Zhang, Gehui Shen, Zhihong Deng. 2020
15 citations recorded by the source corpus.
Lifelong or continual learning remains to be a challenge for artificial neural network, as it is required to be both stable for preservation of old knowledge and plastic for acquisition of new knowledge. It is common to see previous experience get overwritten, which leads to the well-known issue of catastrophic forgetting, especially in the scenario of class-incremental learning (Class-IL). Recently, many lifelong learning methods have been proposed to avoid catastrophic forgetting. However, models which learn without replay of the input data, would encounter another problem which has been ignored, and we refer to it as prior information loss (PIL). In training procedure of Class-IL, as the model has no knowledge about following tasks, it would only extract features necessary for tasks learned so far, whose information is insufficient for joint classification. In this paper, our empirical results on several image datasets show that PIL limits the performance of current state-of-the-art method for Class-IL, the orthogonal weights modification (OWM) algorithm. Furthermore, we propose to combine self-supervised learning, which can provide effective representations without requiring labels, with Class-IL to partly get around this problem. Experiments show superiority of proposed method to OWM, as well as other strong baselines.
Subfields: Artificial Intelligence
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Yilun Du
- Igor Mordatch
- Bahram Zonooz
- Elahe Arani
- Shuang Li
- Jinwoo Shin
- Prashant Bhat
- Antonio Torralba
Citations in the corpus
- Task-Aware Information Routing from Common Representation Space in Lifelong Learning (2023, background): “, 2021b) and prior information loss (Zhang et al., 2020) resulting in aggravated forgetting of previous tasks.”
- Co2L: Contrastive Continual Learning (2021, background): “Concurrent works [43, 15, 30] exploit self-supervised learning objectives to learn more generalizable representations than ones trained with supervised learning objectives.”
- Striking a Balance between Stability and Plasticity for Class-Incremental Learning (2021, background): “This differs from [43] which learns prior information to aid CIL with simple rotation prediction layers.”
- Energy-Based Models for Continual Learning (2020, methodology): “For MAS, we use the result from (Prabhu et al., 2020; Zhang et al., 2020).”
- Task Agnostic Representation Consolidation: a Self-supervised based Continual Learning Approach (2022, background): “Furthermore, the optimization objective in cross-entropy loss encourages learning of representations optimal for the current task sidelining the representations that might be necessary for the future tasks, resulting in prior information loss (Zhang et al., 2020).”