arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 1908.04211cs.CLcs.LG
收藏

关于Transformers的可辨识性

On Identifiability in Transformers

ETH Zurich(苏黎世联邦理工学院);Google Research(谷歌研究院)

Gino Brunner, Yang Liu, Damián Pascual, Oliver Richter, Massimiliano Ciaramita, Roger Wattenhofer

摘要:

在本文中,我们深入研究了Transformer架构,通过研究其两个核心组件:自我注意力和上下文嵌入。特别是,我们研究了注意力权重和令牌嵌入的可识别性,以及将上下文聚合为隐藏令牌。我们发现,序列长度超过注意头尺寸,注意权重是不可识别的。我们提出有效注意作为一种补充工具,以改善解释性解释的基础上注意。此外,我们表明,输入标记在很大程度上保留了它们在整个模型中的身份。我们还发现证据表明,身份信息主要是在嵌入的角度编码,并随着深度逐渐减少。最后,我们展示了强混合的输入信息的生成上下文嵌入通过一种新的量化方法的基础上梯度属性。总的来说,我们表明,自我注意力分布是不能直接解释和目前的工具,以更好地理解和进一步研究Transformer模型。

英文摘要:

In this paper we delve deep in the Transformer architecture by investigating two of its core components: self-attention and contextual embeddings. In particular, we study the identifiability of attention weights and token embeddings, and the aggregation of context into hidden tokens. We show that, for sequences longer than the attention head dimension, attention weights are not identifiable. We propose effective attention as a complementary tool for improving explanatory interpretations based on attention. Furthermore, we show that input tokens retain to a large degree their identity across the model. We also find evidence suggesting that identity information is mainly encoded in the angle of the embeddings and gradually decreases with depth. Finally, we demonstrate strong mixing of input information in the generation of contextual embeddings by means of a novel quantification method based on gradient attribution. Overall, we show that self-attention distributions are not directly interpretable and present tools to better understand and further investigate Transformer 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。

↑