DEFT: Data Efficient Fine-Tuning for Pre-Trained Language Models via Unsupervised Core-Set Selection
This addresses the data efficiency challenge for practitioners fine-tuning language models, though it is incremental as it builds on existing fine-tuning and core-set selection methods.
The paper tackles the problem of reducing data requirements for fine-tuning pre-trained language models by introducing DEFT-UCS, a framework that uses unsupervised core-set selection to identify representative subsets. The result shows that DEFT-UCS models achieve accuracy comparable to the state-of-the-art CoEDIT model across eight datasets and six editing tasks while using 70% less data.
Recent advances have led to the availability of many pre-trained language models (PLMs); however, a question that remains is how much data is truly needed to fine-tune PLMs for downstream tasks? In this work, we introduce DEFT-UCS, a data-efficient fine-tuning framework that leverages unsupervised core-set selection to identify a smaller, representative dataset that reduces the amount of data needed to fine-tune PLMs for downstream tasks. We examine the efficacy of DEFT-UCS in the context of text-editing LMs, and compare to the state-of-the art text-editing model, CoEDIT. Our results demonstrate that DEFT-UCS models are just as accurate as CoEDIT, across eight different datasets consisting of six different editing tasks, while finetuned on 70% less data.