Continual Learning with Large Language Models: Adapting to Concept Drift and New Data Streams

EasyChair Preprint no. 12275

Abstract

This paper addresses the pressing need to adapt LLMs to evolving data distributions and integrate new data streams seamlessly. This work contributes to the advancement of continual learning techniques for LLMs, paving the way for more robust and adaptive natural language understanding systems in dynamic environments. Experimental results on various language understanding tasks demonstrate the effectiveness of our approach in preserving performance on previous tasks while rapidly adapting to changes in the data distribution and accommodating new data streams. Continual learning with large language models (LLMs) presents a formidable challenge due to the dynamic nature of natural language and the emergence of concept drift over time.

Keyphrases: language, large, models

@Booklet{EasyChair:12275,
  author = {Kurez Oroy and Julia Evan},
  title = {Continual Learning with Large Language Models: Adapting to Concept Drift and New Data Streams},
  howpublished = {EasyChair Preprint no. 12275},

  year = {EasyChair, 2024}}