Dynamic Memory Networks for Visual and textual question answering Caiming xiong, Stephen merity, Richard Socher (Submitted on 4 Mar 2016) neural network architectures with MEM Ory and attention mechanisms exhibit certain reasoning capabilities for the for required question. One such architecture, the Dynamic Memory Network (DMN), obtained high accuracy on a variety of the tasks. However, it is not shown whether the architecture achieves strong results for question answering when supporting facts AR E not marked during training or whether it could is applied to other modalities such as images. Based on the DMN, we are propose several improvements to its memory and input modules. Together with the changes we introduce a novel input module for images into order to is able to answer visual questions. Our new dmn+ model improves the "state of" art on both the Visual question answering dataset and the \babi-10k text ques Tion-answering DataSet without supporting fact supervision.
Subjects: |
Neural and Evolutionary Computing (cs.ne); Computation and Language (CS. CL); Computer Vision and Pattern recognition (CS. CV) |
Cite as: |
arxiv:1603.01417 [Cs.ne] |
|
(or Arxiv:1603.01417v1 [cs.ne] for this version) |
Submission HistoryFrom:richard socher [view email]
[V1]Fri, 4 Mar 2016 10:40:28 GMT (2059kb,d)
Which authors of this paper are endorsers? | Disable Mathjax (What is Mathjax?)