Continual Sequence Generation with Adaptive Compositional Modules
Preprint
Abstract
Continual learning is essential for real-world deployment when there is a need to quickly adapt the model to new tasks without forgetting knowledge of old tasks. Existing work on continual sequence generation either always reuses existing parameters to learn new tasks, which is vulnerable to catastrophic forgetting on dissimilar tasks, or blindly adds new parameters for every new task, which could prevent knowledge sharing between similar tasks. To get the best of both worlds, in this work, we propose continual sequence generation with adaptive compositional modules to adaptively add modules in transformer architectures and compose both old and new modules for new tasks. We also incorporate pseudo experience replay to facilitate knowledge transfer in those shared modules. Experiment results on various sequences of generation tasks show that our framework can adaptively add modules or reuse modules based on task similarity, outperforming state-of-the-art baselines in terms of both performance and parameter efficiency. We make our code public at https://github.com/GT-SALT/Adaptive-Compositional-Modules.Keywords
All Related Versions
This publication has 26 references indexed in Scilit:
- An Empirical Survey of Data Augmentation for Limited Data Learning in NLPTransactions of the Association for Computational Linguistics, 2023
- Back to Reality: Weakly-supervised 3D Object Detection with Shape-guided Label Enhancement2022
- Always Be Dreaming: A New Approach for Data-Free Class-Incremental Learning2021
- Memory-Efficient Semi-Supervised Continual Learning: The World is its Own Replay Buffer2021
- Adaptive Adapters: An Efficient Way to Incorporate BERT Into Neural Machine TranslationIEEE/ACM Transactions on Audio, Speech, and Language Processing, 2021
- BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension2020
- Continual Relation Learning via Episodic Memory Activation and Reconsolidation2020
- MixText: Linguistically-Informed Interpolation of Hidden Space for Semi-Supervised Text Classification2020
- PackNet: Adding Multiple Tasks to a Single Network by Iterative Pruning2018
- Deep Residual Learning for Image Recognition2016