Do explanations make VQA models more predictable to a human?
Do explanations make VQA models more predictable to a human?
复制标题
解释是否能让 VQA 模型对人类来说更具可预测性？
DOI：
10.18653/v1/d18-1128
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
Devi Parikh
中科院分区：
文献类型：
--
作者：
Arjun Chandrasekaran;Viraj Prabhu;Deshraj Yadav;Prithvijit Chattopadhyay;Devi Parikh
一系列丰富的研究试图通过生成人类可解释的决策过程“解释”来使深度神经网络变得更加透明，特别是对于视觉问答（VQA）等交互式任务。在这项工作中，我们分析现有的解释是否确实使 VQA 模型（其响应和失败）对人类来说更具可预测性。令人惊讶的是，我们发现他们没有。另一方面，我们发现人机交互方法将模型视为黑匣子。
A rich line of research attempts to make deep neural networks more transparent by generating human-interpretable ‘explanations’ of their decision process, especially for interactive tasks like Visual Question Answering (VQA). In this work, we analyze if existing explanations indeed make a VQA model — its responses as well as failures — more predictable to a human. Surprisingly, we find that they do not. On the other hand, we find that human-in-the-loop approaches that treat the model as a black-box do.