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 2311.07682cs.CLcs.AIcs.LG
收藏

融合以遗忘:通过模型融合减少偏差与选择性记忆

Fuse to Forget: Bias Reduction and Selective Memorization through Model Fusion

UNC Chapel Hill(北卡罗来纳大学教堂山分校);IBM Research(IBM研究院)

Kerem Zaman, Leshem Choshen, Shashank Srivastava

AI总结:

本研究探讨模型融合的逆问题,发现在分类和生成任务中融合能增强模型间共享知识并遗忘非共享知识,从而有效减少微调语言模型中的捷径学习、社会偏差和训练数据记忆,展示了其作为去偏差工具和解决隐私问题的潜力。

摘要:

模型融合研究旨在通过组合权重聚合多个独立模型的知识以提升性能。在本研究中,我们探讨其逆问题:模型融合能否用于减少不需要的知识。我们在三个场景中研究了模型融合的效果:微调语言模型中捷径学习、社会偏差和训练数据记忆。通过涵盖分类和生成任务的实验,我们的分析表明,模型间的共享知识在融合中得到增强,而非共享知识通常被遗忘。基于这一观察,我们展示了模型融合作为去偏差工具的潜力,并展示了其在解决与语言模型相关的隐私问题方面的有效性。

英文摘要:

Model fusion research aims to aggregate the knowledge of multiple individual models to enhance performance by combining their weights. In this work, we study the inverse problem: investigating whether model fusion can be used to reduce unwanted knowledge. We investigate the effects of model fusion in three scenarios: the learning of shortcuts, social biases, and memorization of training data in fine-tuned language models. Through experiments covering classification and generation tasks, our analysis highlights that shared knowledge among models is enhanced during model fusion, while unshared knowledge is usually forgotten. Based on this observation, we demonstrate the potential of model fusion as a debiasing tool and showcase its efficacy in addressing privacy concerns associated with language models.

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。

↑