arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1909.04547cs.CL
收藏

深度网络喜欢读什么?

What do Deep Networks Like to Read?

Technische Universität Darmstadt(达姆施塔特工业大学);DeepMind(深度思维);Insight Centre, NUI Galway(爱尔兰国立大学戈尔韦分校洞察中心);Aylien Ltd.(Aylien有限公司);New York University(纽约大学)

Jonas Pfeiffer, Aishwarya Kamath, Iryna Gurevych, Sebastian Ruder

摘要:

最近对理解神经网络的研究以自上而下的方式探测模型,但只能识别先验已知的模型趋势。我们提出了敏感性识别通过微调(SIFT),一种新的抽象方法,发现一个模型的偏好,而不施加任何先验。通过使用来自固定分类器的梯度微调自动编码器,我们能够以自下而上的方式提取表征不同类型分类器的倾向。我们进一步利用SIFT架构来重组句子,以预测地面真值标签的对立类别,从而揭示固定分类模型中编码的潜在伪影。我们评估我们的方法在三个不同的任务与四个不同的模型。我们对比了模型的倾向,并再现了文献中报道的伪影。

英文摘要:

Recent research towards understanding neural networks probes models in a top-down manner, but is only able to identify model tendencies that are known a priori. We propose Susceptibility Identification through Fine-Tuning (SIFT), a novel abstractive method that uncovers a model's preferences without imposing any prior. By fine-tuning an autoencoder with the gradients from a fixed classifier, we are able to extract propensities that characterize different kinds of classifiers in a bottom-up manner. We further leverage the SIFT architecture to rephrase sentences in order to predict the opposing class of the ground truth label, uncovering potential artifacts encoded in the fixed classification model. We evaluate our method on three diverse tasks with four different models. We contrast the propensities of the models as well as reproduce artifacts reported in the literature.

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。

↑