arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1810.12366cs.AIcs.CLcs.CV
收藏

解释是否使VQA模型对人类来说更可预测?

Do Explanations make VQA Models more Predictable to a Human?

Georgia Institute of Technology(佐治亚理工学院);Facebook AI Research(脸书人工智能研究院)

Arjun Chandrasekaran, Viraj Prabhu, Deshraj Yadav, Prithvijit Chattopadhyay, Devi Parikh

摘要:

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,我们分析了现有的解释是否确实使VQA模型-它的响应以及失败-对人类更可预测。令人惊讶的是,我们发现他们没有。另一方面,我们发现将模型视为黑盒的人在回路方法确实如此。

英文摘要:

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.

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。

↑