arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2306.04067cs.CL
收藏

参数高效方法用于预训练语言模型去偏的实证分析

An Empirical Analysis of Parameter-Efficient Methods for Debiasing Pre-Trained Language Models

University of Oxford(牛津大学);Vienna University of Technology(维也纳工业大学)

Zhongbin Xie, Thomas Lukasiewicz

AI总结:

本文实证评估 prefix tuning、prompt tuning 与 adapter tuning 结合 CDA 的去偏效果,发现其在性别偏见上有效且更省资源,但对种族和宗教偏见作用有限。

摘要:

现代预训练语言模型的规模日益增大,这不仅使其从训练语料中继承了更多类似人类的偏见,也使得缓解此类偏见的计算成本高昂。本文研究了近期的参数高效方法与反事实数据增强(CDA)相结合以缓解偏见的效果。我们在不同语言模型和偏见类型上,使用 prefix tuning、prompt tuning 和 adapter tuning 开展了大量实验,以评估它们的去偏性能以及保留预训练模型内部知识的能力。我们发现,参数高效方法:(i)能有效缓解性别偏见,其中 adapter tuning 始终最为有效,而 prompt tuning 相比 BERT 更适合 GPT-2;(ii)在涉及种族和宗教偏见时效果较差,这可能归因于 CDA 的局限;(iii)能够取得与全量微调相似甚至有时更好的表现,同时提升时间和内存效率,并通过事实检索和下游微调评估表明,可保持 BERT 和 GPT-2 中的内部知识。

英文摘要:

The increasingly large size of modern pretrained language models not only makes them inherit more human-like biases from the training corpora, but also makes it computationally expensive to mitigate such biases. In this paper, we investigate recent parameter-efficient methods in combination with counterfactual data augmentation (CDA) for bias mitigation. We conduct extensive experiments with prefix tuning, prompt tuning, and adapter tuning on different language models and bias types to evaluate their debiasing performance and abilities to preserve the internal knowledge of a pre-trained model. We find that the parameter-efficient methods (i) are effective in mitigating gender bias, where adapter tuning is consistently the most effective one and prompt tuning is more suitable for GPT-2 than BERT, (ii) are less effective when it comes to racial and religious bias, which may be attributed to the limitations of CDA, and (iii) can perform similarly to or sometimes better than full fine-tuning with improved time and memory efficiency, as well as maintain the internal knowledge in BERT and GPT-2, evaluated via fact retrieval and downstream fine-tuning.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑