Mixture of Linear Models Co-supervised by Deep Neural Networks
Mixture of Linear Models Co-supervised by Deep Neural Networks
复制标题
DOI：
10.1080/10618600.2022.2107533
复制
发表时间：
2021-08
影响因子：
2.4
通讯作者：
Beomseok Seo;Lin Lin-Lin;Jia Li
中科院分区：
文献类型：
--
作者：
Beomseok Seo;Lin Lin-Lin;Jia Li
深度神经网络（DNN）已被证明在广泛的应用中实现无与伦比的预测精度。尽管DNN具有强大的性能，但在某些领域，DNN的使用由于其黑箱性质而遇到了阻力。在这篇文章中，我们提出了一种新的方法来估计回归或分类的混合线性模型（MLM），这种方法相对容易解释。我们使用DNN作为最佳预测函数的代理，以便可以有效地估计MLM。我们提出了可视化的方法和定量的方法来解释预测MLM。实验表明，新方法允许我们权衡可解释性和准确性。在经过训练的DNN指导下估计的MLM填补了高度可解释的线性统计模型和高度准确但难以解释的预测因子之间的差距。本文的补充材料可在网上查阅。
ABSTRACT Deep neural networks (DNN) have been demonstrated to achieve unparalleled prediction accuracy in a wide range of applications. Despite its strong performance, in certain areas, the usage of DNN has met resistance because of its black-box nature. In this article, we propose a new method to estimate a mixture of linear models (MLM) for regression or classification that is relatively easy to interpret. We use DNN as a proxy of the optimal prediction function such that MLM can be effectively estimated. We propose visualization methods and quantitative approaches to interpret the predictor by MLM. Experiments show that the new method allows us to tradeoff interpretability and accuracy. The MLM estimated under the guidance of a trained DNN fills the gap between a highly explainable linear statistical model and a highly accurate but difficult to interpret predictor. Supplementary materials for this article are available online.