Rational LAMOL: A Rationale-based Lifelong Learning Framework

By Kasidis Kanwatchara, Thanapapas Horsuwan, Piyawat Lertvittayakumjorn, Boonserm Kijsirikul, Peerapon Vateekul. 2021

23 citations recorded by the source corpus.

Lifelong learning (LL) aims to train a neural network on a stream of tasks while retaining knowledge from previous tasks. However, many prior attempts in NLP still suffer from the catastrophic forgetting issue, where the model completely forgets what it just learned in the previous tasks. In this paper, we introduce Rational LAMOL, a novel end-to-end LL framework for language models. In order to alleviate catastrophic forgetting, Rational LAMOL enhances LAMOL, a recent LL model, by applying critical freezing guided by human rationales. When the human rationales are not available, we propose exploiting unsupervised generated rationales as substitutions. In the experiment, we tested Rational LAMOL on permutations of three datasets from the ERASER benchmark. The results show that our proposed framework outperformed vanilla LAMOL on most permutations. Furthermore, unsupervised rationale generation was able to consistently improve the overall LL performance from the baseline without relying on human-annotated rationales.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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

1. Xisen Jin
2. Chengwei Qin
3. Xiang Ren
4. Jianguo Sun
5. Yongbin Li
6. Henghui Zhu
7. Chang Gao
8. Andrew O. Arnold

In Computer Vision and Pattern Recognition

1. Chengwei Qin
2. Xiang Ren
3. Jianguo Sun
4. Yongbin Li
5. Chang Gao
6. Ying Zhao
7. Xisen Jin
8. Dejiao Zhang

Citations in the corpus

• LFPT5: A Unified Framework for Lifelong Few-shot Language Learning Based on Prompt Tuning of T5 (2021, background): “These include LAMOL (Sun et al., 2019) and its improvements (Chuang et al., 2020; Sun et al., 2020; Kanwatchara et al., 2021).”
• Prompt Conditioned VAE: Enhancing Generative Replay for Lifelong Learning in Task-Oriented Dialogue (2022, background): “Kanwatchara et al. (2021) freeze critical parameters in LAMOL based on rationales, but those rationales are not always available for NLP tasks.”
• Lifelong Pretraining: Continually Adapting Language Models to Emerging Corpora (2021, background): “(Sun et al., 2020; Kanwatchara et al., 2021)).”