sigconf.tex

\documentclass[sigconf]{acmart}

\usepackage{booktabs} % For formal tables
\usepackage{subcaption}

% Copyright
%\setcopyright{none}
%\setcopyright{acmcopyright}
%\setcopyright{acmlicensed}
\setcopyright{rightsretained}
%\setcopyright{usgov}
%\setcopyright{usgovmixed}
%\setcopyright{cagov}
%\setcopyright{cagovmixed}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% FOR ADDING COLORED COMMENTS
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\usepackage{xcolor}
\usepackage{ifthen}
%------- Tool to display colored notes and changes in the draft and remove them easily
% usage: cp-paste these two lines into the preamble of your document
% \input{markupdraft.tex}  % or cp-paste the below into the preamble
% \renewcommand{\markupdraft}[2]{}  % remove all todo's, notes and coloring of changes
                                    % like for the final version
% The available commands are \del, \new, \nnew, \note, \nnote, \todo (same as \TODO).
% All commands take as optional argument the author.  The nn versions give more pronounced colors. 
% Example:
%   This is\del{ an experimental version of}\note[Niko]{remark the space within the del command} 
%   a \new{simple to use} LaTeX markup.\todo{improve this example} 
%------- Meta-command to markup notes and changes and *remove those markups easily*
\newcommand{\markupdraft}[2]{% {#1: {color|display} command}{#2: desired color or text}
%  the next lines can be incommented, if respectively certain notes or coloring should disappear
    \ifthenelse{\equal{#1}{display}}{#2}{}%                 % display only in draft version
    \ifthenelse{\equal{#1}{color}}{\color{#2}}{}%           % colored only in draft (for \new command)
}
% ------ helper commands using \markupdraft, author is optional argument  
\newcommand{\notecolored}[3][]{\markupdraft{display}{{\color{#2}\noindent[Note (#1): #3]}}}
\newcommand{\newcolored}[3][]{{\markupdraft{color}{#2}#3}%  % kept in the final print
    \ifthenelse{\equal{#1}{}}{}{\markupdraft{display}{{\color{yellow!70!black}[#1]}}}} 

% ------ end user markupdraft-commands, optional argument is the author: 
\newcommand{\del}[2][]{{\markupdraft{display}{{\color{orange}[removed: ``#2''[#1]]}}}} % (to be) removed
\newcommand{\new}[2][]{\newcolored[#1]{blue}{#2}}%  % kept in the final print
\newcommand{\nnew}[2][]{\newcolored[#1]{red}{#2}}%  % kept in the final print
\newcommand{\note}[2][]{\notecolored[#1]{green}{#2}}    
\newcommand{\nnote}[2][]{\notecolored[#1]{magenta}{#2}}   % note with Anne's preferred default color
\newcommand{\future}[2][]{{\markupdraft{display}{{\color{magenta}[[future:]#2[#1]]}}}}  % don't keep in current final print, but for the future 
\newcommand{\TODO}[2][]{\markupdraft{display}{{\color{red}~\\\noindent== TODO: #2 {\color{yellow}(#1)} ==\\}}}
\newcommand{\todo}[2][]{\markupdraft{display}{{\color{red}\noindent++TODO: #2 {\color{yellow}(#1)}++}}}
% ------ Final version
%\renewcommand{\del}[2]{} % make the removed part invisible
%\renewcommand{\markupdraft}[2]{}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% END ADDING COLORED COMMENTS
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\newcommand{\shin}[1]{\note[Shinichi]{\color{magenta} #1}}
\newcommand{\suga}[1]{\note[Suganuma]{\color{cyan} #1}}
\newcommand{\usrc}[1]{\note[UserC]{\color{green} #1}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


%% DOI
%\acmDOI{10.475/123_4}
%
%% ISBN
%\acmISBN{123-4567-24-567/08/06}
%
%%Conference
\acmConference[GECCO '17]{the Genetic and Evolutionary Computation Conference 2017}{July 15--19, 2017}{Berlin, Germany}
\acmYear{2017}
\copyrightyear{2017}

%\acmPrice{15.00}



\begin{document}
%\title{A Genetic Programming Approach to Designing Convolutional Neural Network Architectures}
\title[A Genetic Programming Approach to Designing CNN Architectures]{A Genetic Programming Approach to Designing Convolutional Neural Network Architectures}

%\author{Masanori Suganuma}
%\affiliation{%
%  \institution{Graduate school of Environment and Information Sciences, Yokohama National University}
%  \streetaddress{79-7 Tokiwadai Hodogaya-ku}
%  \city{Yokohama} 
%  \country{Japan} 
%  \postcode{240-8501}
%}
%\email{suganuma-masanori-hf@ynu.jp}
%%
%\author{Shinichi Shirakawa}
%\affiliation{%
%  \institution{Graduate school of Environment and Information Sciences, Yokohama National University}
%  \streetaddress{79-7 Tokiwadai Hodogaya-ku}
%  \city{Yokohama} 
%  \country{Japan} 
%  \postcode{240-8501}
%}
%\email{shirakawa-shinich-bg@ynu.ac.jp}
%%
%\author{Tomoharu Nagao}
%\affiliation{%
%  \institution{Graduate school of Environment and Information Sciences, Yokohama National University}
%  \streetaddress{79-7 Tokiwadai Hodogaya-ku}
%  \city{Yokohama} 
%  \country{Japan} 
%  \postcode{240-8501}
%}
%\email{nagao@ynu.ac.jp}

\author{1st Autor}
\affiliation{%
  \institution{Institution line}
  \streetaddress{Street Adress}
  \city{City} 
  \country{Country} 
  \postcode{Post Code}
}
\email{email adress}
%
\author{2nd Autor}
\affiliation{%
  \institution{Institution line}
  \streetaddress{Street Adress}
  \city{City} 
  \country{Country} 
  \postcode{Post Code}
}
\email{email adress}
%
\author{3rd Autor}
\affiliation{%
  \institution{Institution line}
  \streetaddress{Street Adress}
  \city{City} 
  \country{Country} 
  \postcode{Post Code}
}
\email{email adress}

% The default list of authors is too long for headers}
%\renewcommand{\shortauthors}{B. Trovato et. al.}


\begin{abstract}
The convolutional neural network (CNN), which is one of the deep learning models, has seen much success in a variety of computer vision tasks. However, designing CNN architectures still requires expert knowledge and \new{a lot of trial and error.} In this paper, we attempt to automatically construct \new{CNN} architectures for an image classification task based on Cartesian genetic programming (CGP). In our method, we adopt \new{highly functional} modules\new{,} such as convolutional blocks and tensor concatenation\new{,} as the node functions in CGP. The CNN structure and connectivity represented by the CGP encoding method are optimized to maximize the validation accuracy. To evaluate the proposed method, we constructed \new{a} CNN architecture for the image classification task with the CIFAR-10 dataset. The experimental result shows that the proposed method can \new{be used to} automatically find the competitive CNN architecture compared with state-of-the-art models.
%Deep learning methods such as convolutional neural network(CNN) have seen much success in a variety of machine learning tasks.
%However, designing CNN architectures still requires expert experience and many iterations.
%In this paper, we propose a genetic programming based approach that automatically designs CNN architectures.
%Our method represents the CNN architecture as a feedforward network structure that is similar to cartesian genetic programming, and optimizes the structure and connectivity of the network to maximize the accuracy on a learning set.
%On the image classification task with the CIFAR-10 dataset, our method achieves a test error rate of $8.17$, which is competitive with state-of-the-art-methods.
\end{abstract}

%
% The code below should be generated by the tool at
% http://dl.acm.org/ccs.cfm
% Please copy and paste the code instead of the example below. 
%
\begin{CCSXML}
<ccs2012>
<concept>
<concept_id>10010147.10010178.10010205.10010206</concept_id>
<concept_desc>Computing methodologies~Heuristic function construction</concept_desc>
<concept_significance>500</concept_significance>
</concept>
<concept>
<concept_id>10010147.10010257.10010293.10010294</concept_id>
<concept_desc>Computing methodologies~Neural networks</concept_desc>
<concept_significance>500</concept_significance>
</concept>
<concept>
<concept_id>10010147.10010178.10010224.10010245</concept_id>
<concept_desc>Computing methodologies~Computer vision problems</concept_desc>
<concept_significance>300</concept_significance>
</concept>
</ccs2012>
\end{CCSXML}

\ccsdesc[500]{Computing methodologies~Heuristic function construction}
\ccsdesc[500]{Computing methodologies~Neural networks}
\ccsdesc[300]{Computing methodologies~Computer vision problems}

% We no longer use \terms command
%\terms{Theory}

\keywords{genetic programming, convolutional neural network, designing neural network architectures, deep learning}


\maketitle

\input{body-conf}

\bibliographystyle{ACM-Reference-Format}
\bibliography{sigconf} 

\end{document}