Class-Incremental Learning based on Label Generation
This addresses the challenge of continual learning for NLP practitioners, though it appears incremental as it builds on existing CIL and label generation ideas.
The paper tackles catastrophic forgetting in class-incremental learning for pre-trained language models by reformulating it as a continual label generation problem, resulting in VAG, a method that outperforms baselines by a large margin.
Despite the great success of pre-trained language models, it is still a challenge to use these models for continual learning, especially for the class-incremental learning (CIL) setting due to catastrophic forgetting (CF). This paper reports our finding that if we formulate CIL as a continual label generation problem, CF is drastically reduced and the generalizable representations of pre-trained models can be better retained. We thus propose a new CIL method (VAG) that also leverages the sparsity of vocabulary to focus the generation and creates pseudo-replay samples by using label semantics. Experimental results show that VAG outperforms baselines by a large margin.