Self-Assembling Modular Networks for Interpretable Multi-Hop Reasoning
Self-Assembling Modular Networks for Interpretable Multi-Hop Reasoning
复制标题
用于可解释多跳推理的自组装模块化网络
DOI：
--
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Mohit Bansal
中科院分区：
文献类型：
--
作者：
Yichen Jiang;Mohit Bansal
多跳QA需要一个模型来连接分散在长上下文中的多个证据来回答问题。最近提出的HotpotQA(Yang等人，2018)数据集由体现四种不同多跳推理范式(两个桥实体设置、检查多个属性以及比较两个实体)的问题组成，这使得单个神经网络难以处理所有四个实体。在这项工作中，我们提出了一个用于多跳推理的可解释的、基于控制器的自组装神经模块网络(Hu等人，2017,2018)，其中我们设计了四个新的模块(查找、重定位、比较、NoOp)来执行独特类型的语言推理。基于一个问题，我们的布局控制器RNN动态地推理一系列模块来构建整个网络。实验表明，我们的动态多跳模块化网络比静态的单跳基线(无论是常规评估还是对抗性评估)都取得了显著的改进。我们通过三个分析进一步证明了我们模型的可解释性。首先，控制器可以将多跳问题软分解为多个单跳子问题，以促进主网络的组成推理行为。其次，控制器可以预测符合人类专家设计的布局。最后，中间模块可以通过解决来自控制器的子问题来推断连接两个相距遥远的支持事实的实体。
Multi-hop QA requires a model to connect multiple pieces of evidence scattered in a long context to answer the question. The recently proposed HotpotQA (Yang et al., 2018) dataset is comprised of questions embodying four different multi-hop reasoning paradigms (two bridge entity setups, checking multiple properties, and comparing two entities), making it challenging for a single neural network to handle all four. In this work, we present an interpretable, controller-based Self-Assembling Neural Modular Network (Hu et al., 2017, 2018) for multi-hop reasoning, where we design four novel modules (Find, Relocate, Compare, NoOp) to perform unique types of language reasoning. Based on a question, our layout controller RNN dynamically infers a series of reasoning modules to construct the entire network. Empirically, we show that our dynamic, multi-hop modular network achieves significant improvements over the static, single-hop baseline (on both regular and adversarial evaluation). We further demonstrate the interpretability of our model via three analyses. First, the controller can softly decompose the multi-hop question into multiple single-hop sub-questions to promote compositional reasoning behavior of the main network. Second, the controller can predict layouts that conform to the layouts designed by human experts. Finally, the intermediate module can infer the entity that connects two distantly-located supporting facts by addressing the sub-question from the controller.
DOI：
10.18653/v1/n19-1405
发表时间：
2019
期刊：
--
影响因子：
--
作者：
Jifan Chen;Greg Durrett
通讯作者：
Jifan Chen;Greg Durrett
DOI：
--
发表时间：
2016-11
期刊：
ArXiv
影响因子：
--
作者：
Barret Zoph;Quoc V. Le
通讯作者：
Barret Zoph;Quoc V. Le