Decision-Guided Weighted Automata Extraction from Recurrent Neural Networks
Decision-Guided Weighted Automata Extraction from Recurrent Neural Networks
复制标题
DOI：
10.1609/aaai.v35i13.17391
复制
发表时间：
2021-05
期刊：
影响因子：
--
通讯作者：
Xiyue Zhang;Xiaoning Du;Xiaofei Xie;Lei Ma;Yang Liu;Mengfei Sun
中科院分区：
文献类型：
--
作者：
Xiyue Zhang;Xiaoning Du;Xiaofei Xie;Lei Ma;Yang Liu;Mengfei Sun
经常性的神经网络（RNN）证明了它们在学习和处理顺序数据（例如语音和自然语言）方面的有效性。但是，由于神经网络的黑盒性质，了解RNN的决策逻辑非常具有挑战性。最近已经取得了一些进展，以近似加权自动机的RNN行为。它们提供了更好的解释性，但仍然遭受可扩展性差。在本文中，我们提出了一种新的方法，可以在目标RNN的决策和上下文信息的指导下提取加权自动机。特别是，我们确定了RNN逐步预测决策的模式，以指导自动机状态的形成。此外，我们提出了一种状态组成方法，以增强提取模型的上下文意识。我们对典型的RNN任务的深入评估，包括语言模型和分类，证明了我们方法比最先进的方法的有效性和优势。评估结果表明，即使在大规模任务上，我们的方法也可以实现RNN的准确近似。
Recurrent Neural Networks (RNNs) have demonstrated their effectiveness in learning and processing sequential data (e.g., speech and natural language). However, due to the black-box nature of neural networks, understanding the decision logic of RNNs is quite challenging. Some recent progress has been made to approximate the behavior of an RNN by weighted automata. They provide better interpretability, but still suffer from poor scalability. In this paper, we propose a novel approach to extracting weighted automata with the guidance of a target RNN's decision and context information. In particular, we identify the patterns of RNN's step-wise predictive decisions to instruct the formation of automata states. Further, we propose a state composition method to enhance the context-awareness of the extracted model. Our in-depth evaluations on typical RNN tasks, including language model and classification, demonstrate the effectiveness and advantage of our method over the state-of-the-arts. The evaluation results show that our method can achieve accurate approximation of an RNN even on large-scale tasks.