arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2305.04493cs.CL
收藏

序列到序列模型的词元级拟合问题

Token-Level Fitting Issues of Seq2seq Models

Zhejiang University(浙江大学);Westlake University(西湖大学);Nanyang Technological University(南洋理工大学);Westlake Institute for Advanced Study(西湖高等研究院)

Guangsheng Bao, Zhiyang Teng, Yue Zhang

AI总结:

本文发现早停训练的seq2seq模型存在词元级过拟合与欠拟合并存问题,并识别出词频、词性和预测差异等影响因素。

摘要:

序列到序列(seq2seq)模型已广泛用于自然语言处理、计算机视觉及其他深度学习任务。我们发现,使用早停法训练的seq2seq模型在词元级别上存在问题。具体而言,当训练停止时,词汇表中的某些词元表现出过拟合,而其他词元则欠拟合。实验表明,这种现象在不同模型中普遍存在,甚至在微调的大型预训练模型中也是如此。我们确定了影响词元级拟合的三个主要因素,包括词元频率、词性和预测差异。此外,我们发现语言、模型规模、领域、数据规模和预训练等外部因素也会影响词元的拟合。

英文摘要:

Sequence-to-sequence (seq2seq) models have been widely used for natural language processing, computer vision, and other deep learning tasks. We find that seq2seq models trained with early-stopping suffer from issues at the token level. In particular, while some tokens in the vocabulary demonstrate overfitting, others underfit when training is stopped. Experiments show that the phenomena are pervasive in different models, even in fine-tuned large pretrained-models. We identify three major factors that influence token-level fitting, which include token frequency, parts-of-speech, and prediction discrepancy. Further, we find that external factors such as language, model size, domain, data scale, and pretraining can also influence the fitting of tokens.

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。

↑