研究Transformer对序列长度的过拟合问题，揭示其在长序列上表现差但部分正确。
Adversarial Testing as a Tool for Interpretability: Length-based Overfitting of Elementary Functions in Transformers
- 用编辑函数和错误指标分析Transformer行为
- 短序列可泛化，长序列严重过拟合但有部分正确输出
- 结构特征如子段长度影响模型决策，适合关注可解释性的研究者
Transformer模型倾向于过拟合训练数据的多个方面，例如整体序列长度。本文通过一组定义好的错误指标，研究了序列到序列Transformer对基本字符串编辑函数的行为。结果显示，模型通常能泛化到更短的序列，但对更长的序列则表现不佳，尽管常能得到部分正确的答案。此外，我们发现序列的其他结构性特征（如子段长度）可能同样重要。我们假设，模型在学习任务的算法性特征的同时，也学习了结构性特征，但当二者冲突时，模型往往优先遵循结构性特征。
原文摘要 · Abstract (English)
The Transformer model has a tendency to overfit various aspects of the training data, such as the overall sequence length. We study elementary string edit functions using a defined set of error indicators to interpret the behaviour of the sequence-to-sequence Transformer. We show that generalization to shorter sequences is often possible, but confirm that longer sequences are highly problematic, although partially correct answers are often obtained. Additionally, we find that other structural characteristics of the sequences, such as subsegment length, may be equally important. We hypothesize that the models learn algorithmic aspects of the tasks simultaneously with structural aspects but adhering to the structural aspects is unfortunately often preferred by Transformer when they come into conflict.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。