arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2203.16455cs.LG
收藏

通过对偶性明确深度神经网络的隐式不可解释性

Explicitising The Implicit Intrepretability of Deep Neural Networks Via Duality

Indian Institute of Technology Madras(印度理工学院马德拉斯分校)

Chandrashekar Lakshminarayanan, Amit Vikram Singh, Arun Rajkumar

摘要:

Lakshminarayanan和Singh [2020]最近的工作为具有整流线性单元(ReLU)的完全连接深度神经网络(DNN)提供了双重视图。结果表明,(i)门中的信息在分析上由称为神经路径内核(NPK)的内核表征,以及(ii)门中学习了最关键的信息,因为,给定学习的门,权重可以从头开始重新训练,而不会显着降低性能。在本文中,我们使用的双重观点,重新思考传统的DNN解释,从而明确了DNN的隐含的可解释性。为此,我们首先展示了新的理论性质,即分别存在卷积层和跳跃连接时NPK的旋转不变性和系综结构。我们的理论导致了两个令人惊讶的经验结果,挑战了传统的智慧:(i)即使使用恒定的1输入,也可以训练权重,(ii)门控掩码可以被打乱,而不会有任何显着的性能损失。这些结果激发了一类新的网络,我们称之为深度线性门控网络(DLGN)。使用双重提升现象的DLGN为DNN的更直接和更简单的解释铺平了道路,而不是传统的解释。我们通过对CIFAR-10和CIFAR-100的大量实验表明,这些DLGN导致更好的可解释性-准确性权衡。

英文摘要:

Recent work by Lakshminarayanan and Singh [2020] provided a dual view for fully connected deep neural networks (DNNs) with rectified linear units (ReLU). It was shown that (i) the information in the gates is analytically characterised by a kernel called the neural path kernel (NPK) and (ii) most critical information is learnt in the gates, in that, given the learnt gates, the weights can be retrained from scratch without significant loss in performance. Using the dual view, in this paper, we rethink the conventional interpretations of DNNs thereby explicitsing the implicit interpretability of DNNs. Towards this, we first show new theoretical properties namely rotational invariance and ensemble structure of the NPK in the presence of convolutional layers and skip connections respectively. Our theory leads to two surprising empirical results that challenge conventional wisdom: (i) the weights can be trained even with a constant 1 input, (ii) the gating masks can be shuffled, without any significant loss in performance. These results motivate a novel class of networks which we call deep linearly gated networks (DLGNs). DLGNs using the phenomenon of dual lifting pave way to more direct and simpler interpretation of DNNs as opposed to conventional interpretations. We show via extensive experiments on CIFAR-10 and CIFAR-100 that these DLGNs lead to much better interpretability-accuracy tradeoff.

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。

↑