arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2410.13641cs.CL
收藏

一种面向大语言模型包容性生成的主动学习框架

An Active Learning Framework for Inclusive Generation by Large Language Models

School of Computing and Information, University of Pittsburgh(匹兹堡大学计算与信息学院);Khoury College of Computer Science, Northeastern University(东北大学科里计算机学院)

Sabit Hassan, Anthony Sicilia, Malihe Alikhani

AI总结:

提出一种结合聚类与知识蒸馏的主动学习框架,首次实现生成任务的有效主动学习,在反叙事和风格迁移中提升模型性能2%-10%,增强多样性与鲁棒性。

摘要:

确保大语言模型(LLMs)生成的文本能够代表多样化的子群体至关重要,尤其是当与代表性不足群体相关的关键概念在训练数据中稀缺时。我们通过一种新颖的基于聚类的主动学习框架来解决这一挑战,该框架结合了知识蒸馏。所提出的框架对学习模型的中间输出进行变换,从而首次实现了生成任务的有效主动学习。聚类与知识蒸馏的整合,使得模型在没有底层数据分布先验知识和过度人力投入的情况下,更具代表性。我们通过反叙事和风格迁移的案例研究在实践中验证了我们的方法。我们与模型训练同步构建了两个新数据集,显示出相比基线模型2%-10%的性能提升。我们的结果还表明,模型在不同数据子群体上的性能更加一致,且词汇多样性增加,这凸显了我们的模型对现有数据偏斜的鲁棒性。此外,我们的结果显示,通过我们的方法获取的数据提高了未参与学习循环的辅助模型的性能,展示了该框架的实际效用。

英文摘要:

Ensuring that Large Language Models (LLMs) generate text representative of diverse sub-populations is essential, particularly when key concepts related to under-represented groups are scarce in the training data. We address this challenge with a novel clustering-based active learning framework, enhanced with knowledge distillation. The proposed framework transforms the intermediate outputs of the learner model, enabling effective active learning for generative tasks for the first time. Integration of clustering and knowledge distillation yields more representative models without prior knowledge of underlying data distribution and overbearing human efforts. We validate our approach in practice through case studies in counter-narration and style transfer. We construct two new datasets in tandem with model training, showing a performance improvement of 2%-10% over baseline models. Our results also show more consistent performance across various data subgroups and increased lexical diversity, underscoring our model's resilience to skewness in available data. Further, our results show that the data acquired via our approach improves the performance of secondary models not involved in the learning loop, showcasing practical utility of the framework.

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。

↑