Modern deep learning models are notoriously opaque, which has motivated the
development of methods for interpreting how deep models predict. This goal is
usually approached with attribution method, which assesses
通过分析从基于提示的模型中提取的归因得分的合理性和忠实性,并将其与从微调模型和大型语言模型中提取的归因得分进行比较,我们发现使用基于提示的范例(无论是基于编码器的模型还是解码器的模型)比在低资源环境下微调模型产生更合理的解释,并且 Shapley Value Sampling 在产生更合理和忠实的解释方面始终优于注意力和积分梯度。