-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathAbstract.tex
105 lines (90 loc) · 4.19 KB
/
Abstract.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
\documentclass[uplatex,a4paper,twocolumn,8pt,dvipdfmx]{jsarticle}
\usepackage[dvipdfmx]{graphicx}
\usepackage{diagbox} % 旧\usepackage{slashbox}
\usepackage{wrapfig}
\usepackage{setspace}
\usepackage{latexsym}
\usepackage{tabularx} % Table用
\usepackage{multicol} % セル結合用
\usepackage{multirow} % セル結合用
% \AtBeginDvi{\special{pdf:mapfile msmingoth.map}} % WindowsでMS明朝を使うとき用
\usepackage[top=15truemm,bottom=20truemm,left=16truemm,right=16truemm]{geometry}
\newcolumntype{C}[1]{>{\centering\arraybackslash}p{#1}}
\newcolumntype{L}[1]{>{\raggedright\arraybackslash}p{#1}}
\newcolumntype{R}[1]{>{\raggedleft\arraybackslash}p{#1}}
\title{uplatexを使った2段組文書作成} %タイトル
\author{Haruka \thanks{国立高等専門学校 全知全能科}}
\date{2021年11月26日}
\begin{document}
\maketitle
%タイトル
%本文
\section{はじめに}
最近は Over Leaf 等のオンラインの\LaTeX サービスがあるので,
それらを使うといいと思うが画像のコンパイル等が重くなるので
やっぱりローカルで環境を作ったほうがいいかもしれない.
upLaTex は Unicode が使えるので記号等の問題が少ないのがメリットかもしれない.
% \vspace{-0.5zh} % ちょっと行間を詰めた
\section{関連研究}
既存研究として,Vinyalsら\cite{Vinyals2015ShowAT}は,
畳み込みニューラルネットワーク(CNN)と
再起型ニューラルネットワーク(RNN)を用いた
エンコーダー・デコーダー構造により,
画像特徴とキャプションから,文の生成を学習する手法を提案している.
\section{提案手法}
Vinyalsらの提案するImage Captioningのエンコーダー・デコーダー構造を参考に,
人間を用いた人力キャプション生成を提案する(図\ref{fig:model}).
\begin{figure}[htbp]
\centering
\includegraphics[width=0.6\linewidth]{fig/human_captioning.jpg}
\caption{絵を見て内容を言語にて回答}
\label{fig:model}
\end{figure}
\section{生成実験}
友達と協力させる人力キャプション生成を実装する(図\ref{fig:ours_fr}).
既存手法としてVinyalsらの手法(NIC)と,人力キャプション生成モデル,
それに友人を導入したと友人協力モデルを比較した.
表\ref{tb:score}に示すように,人力キャプション生成モデルの方が
著しく高いスコアとなった.
また,友人協力モデルは最も高い評価となったものの,
人力キャプション生成モデルと比較すると差はわずかである.
% 表をトップに配置される
\begin{table}[tbp]
\centering
\small
\caption{評価用データセットに対するキャプション生成の品質.
1,2,3,4は$n$-gramsを使ったBleuスコア.
数値が高いほどGround Truthとの類似度が高い.
Ours+Fはモデルと友人(Friend)を相談させるネットワークとする.
}
\begin{tabular}{ L{1.0cm} *{4}{C{0.5cm}} C{0cm} *{3}{C{0.55cm}} }
% \hline
\hline
& \multicolumn{4}{c}{Bleu} && \multicolumn{3}{c}{Bert} \\
\cline{2-5}\cline{7-9}
& 1 & 2 & 3 & 4 && P & R & F1\\ \hline
NIC\cite{Vinyals2015ShowAT} & 0.113 & 0.018 & 0.001 & 0.000 && 0.856 & 0.878 & 0.867\\
Ours & {\bf 1.00} & 0.990 & 0.980 & 0.900 && 0.990 & 0.990 & 0.990\\
Ours+Fr & {\bf 1.00} & {\bf 1.00} & {\bf 0.999} & {\bf 0.990} && {\bf 0.999} & {\bf 0.999} & {\bf 0.999}\\
\hline
\end{tabular}
\label{tb:score}
\end{table}
\begin{figure}[tbp]
\centering
\vspace{0.3cm}
\includegraphics[clip, width=0.8\linewidth]{fig/kids_katawokumu.png}
\caption{友人と協力モデル(Ours+Fr)の図}
\label{fig:ours_fr}
\vspace{-0.5cm}
\end{figure}
\section{まとめ・今後の課題}
完璧すぎてありません.
\renewcommand{\refname}{参考文献}
%% 参考文献
%% アルファベット順(五十音順)
%\bibliographystyle{jplain}
%% 参照順
\bibliographystyle{junsrt}
\bibliography{mybib}
\end{document}