update

Author: imhuay

_notes/Frontiers of Natural Language Processing.pdf

@@ -10,39 +10,39 @@ Index
 <!-- TOC -->
 
 - [VQA 简述](#vqa-简述)
-  - [VQA 与其他图像任务](#vqa-与其他图像任务)
-    - [基于对象检测的任务](#基于对象检测的任务)
-    - [图像描述任务](#图像描述任务)
-      - [DenseCap](#densecap)
-  - [VQA 中的数据集](#vqa-中的数据集)
-    - [DAQUAR](#daquar)
-    - [COCO-QA](#coco-qa)
-    - [VQA Dataset](#vqa-dataset)
-    - [FM-IQA](#fm-iqa)
-    - [Visual Genome](#visual-genome)
-    - [Visual7W](#visual7w)
-    - [SHAPES](#shapes)
-  - [VQA 的评价方法 TODO](#vqa-的评价方法-todo)
+    - [VQA 与其他图像任务](#vqa-与其他图像任务)
+        - [基于对象检测的任务](#基于对象检测的任务)
+        - [图像描述任务](#图像描述任务)
+            - [DenseCap](#densecap)
+    - [VQA 中的数据集](#vqa-中的数据集)
+        - [DAQUAR](#daquar)
+        - [COCO-QA](#coco-qa)
+        - [VQA Dataset](#vqa-dataset)
+        - [FM-IQA](#fm-iqa)
+        - [Visual Genome](#visual-genome)
+        - [Visual7W](#visual7w)
+        - [SHAPES](#shapes)
+    - [VQA 的评价方法 TODO](#vqa-的评价方法-todo)
 - [主流模型与方法](#主流模型与方法)
-  - [基线模型](#基线模型)
-    - [分类模型](#分类模型)
-    - [生成模型](#生成模型)
-  - [贝叶斯模型](#贝叶斯模型)
-  - [基于 Attention 的模型](#基于-attention-的模型)
-    - [基于 Edge Boxes 的方法](#基于-edge-boxes-的方法)
-    - [基于 Uniform Grid 的方法](#基于-uniform-grid-的方法)
-      - [[49] Stacked Attention Networks for Image Question Answering（SAN）](#49-stacked-attention-networks-for-image-question-answeringsan)
-      - [[48] Ask, Attend and Answer: Exploring Question-Guided Spatial Attention for Visual Question Answering](#48-ask-attend-and-answer-exploring-question-guided-spatial-attention-for-visual-question-answering)
-      - [[52] Dynamic memory networks for visual and textual question answering](#52-dynamic-memory-networks-for-visual-and-textual-question-answering)
-      - [[54] Hierarchical Question-Image Co-Attention for Visual Question Answering](#54-hierarchical-question-image-co-attention-for-visual-question-answering)
-      - [[56] Dual attention networks for multimodal reasoning and matching](#56-dual-attention-networks-for-multimodal-reasoning-and-matching)
-  - [基于双线性池化的模型](#基于双线性池化的模型)
-    - [[46] Multimodal compact bilinear pooling for visual question answering and visual grounding](#46-multimodal-compact-bilinear-pooling-for-visual-question-answering-and-visual-grounding)
-    - [[57] Hadamard Product for Low-rank Bilinear Pooling](#57-hadamard-product-for-low-rank-bilinear-pooling)
-  - [组合模型](#组合模型)
-    - [[44] Deep Compositional Question Answering with Neural Module Networks](#44-deep-compositional-question-answering-with-neural-module-networks)
-    - [[55] Training recurrent answering units with joint loss minimization for VQA](#55-training-recurrent-answering-units-with-joint-loss-minimization-for-vqa)
-  - [其他模型 TODO](#其他模型-todo)
+    - [基线模型](#基线模型)
+        - [分类模型](#分类模型)
+        - [生成模型](#生成模型)
+    - [贝叶斯模型](#贝叶斯模型)
+    - [基于 Attention 的模型](#基于-attention-的模型)
+        - [基于 Edge Boxes 的方法](#基于-edge-boxes-的方法)
+        - [基于 Uniform Grid 的方法](#基于-uniform-grid-的方法)
+            - [[49] Stacked Attention Networks for Image Question Answering（SAN）](#49-stacked-attention-networks-for-image-question-answeringsan)
+            - [[48] Ask, Attend and Answer: Exploring Question-Guided Spatial Attention for Visual Question Answering](#48-ask-attend-and-answer-exploring-question-guided-spatial-attention-for-visual-question-answering)
+            - [[52] Dynamic memory networks for visual and textual question answering](#52-dynamic-memory-networks-for-visual-and-textual-question-answering)
+            - [[54] Hierarchical Question-Image Co-Attention for Visual Question Answering](#54-hierarchical-question-image-co-attention-for-visual-question-answering)
+            - [[56] Dual attention networks for multimodal reasoning and matching](#56-dual-attention-networks-for-multimodal-reasoning-and-matching)
+    - [基于双线性池化的模型](#基于双线性池化的模型)
+        - [[46] Multimodal compact bilinear pooling for visual question answering and visual grounding](#46-multimodal-compact-bilinear-pooling-for-visual-question-answering-and-visual-grounding)
+        - [[57] Hadamard Product for Low-rank Bilinear Pooling](#57-hadamard-product-for-low-rank-bilinear-pooling)
+    - [组合模型](#组合模型)
+        - [[44] Deep Compositional Question Answering with Neural Module Networks](#44-deep-compositional-question-answering-with-neural-module-networks)
+        - [[55] Training recurrent answering units with joint loss minimization for VQA](#55-training-recurrent-answering-units-with-joint-loss-minimization-for-vqa)
+    - [其他模型 TODO](#其他模型-todo)
 - [参考文献](#参考文献)
 
 <!-- /TOC -->
@@ -363,7 +363,7 @@ Index
   > 这个过程实际上跟**卷积**本身很像。
   <div align="center"><img src="../_assets/TIM截图20180911153132.png" height="" /></div>
 
-**Attention 的作用**
+<!-- **Attention 的作用** -->
 
 ### 基于 Edge Boxes 的方法
 > [63, 51]