Lifelong Language Knowledge Distillation

TL;DR

This paper introduces Lifelong Language Knowledge Distillation (L2KD), a method that enhances lifelong language learning models by using knowledge distillation to retain past knowledge while adapting to new tasks, reducing performance degradation.

Contribution

L2KD is a simple, effective technique that can be integrated into existing LLL architectures to mitigate knowledge loss during sequential task learning.

Findings

L2KD improves state-of-the-art LLL performance.

It reduces degradation compared to multi-task models.

Effective for sequence generation and text classification.

Abstract

It is challenging to perform lifelong language learning (LLL) on a stream of different tasks without any performance degradation comparing to the multi-task counterparts. To address this issue, we present Lifelong Language Knowledge Distillation (L2KD), a simple but efficient method that can be easily applied to existing LLL architectures in order to mitigate the degradation. Specifically, when the LLL model is trained on a new task, we assign a teacher model to first learn the new task, and pass the knowledge to the LLL model via knowledge distillation. Therefore, the LLL model can better adapt to the new task while keeping the previously learned knowledge. Experiments show that the proposed L2KD consistently improves previous state-of-the-art models, and the degradation comparing to multi-task models in LLL tasks is well mitigated for both sequence generation and text classification tasks.