~

NIPS2020/main.tex

@@ -25,7 +25,7 @@
 \usepackage{amsfonts}       % blackboard math symbols
 \usepackage{nicefrac}       % compact symbols for 1/2, etc.
 \usepackage{microtype}      % microtypography
-
+\usepackage[pdftex]{graphicx}
 
 %%Added by Du, Zidong
 \usepackage{ifthen}
@@ -168,17 +168,56 @@ systems. However, even using the multi-agent systems, previous works at root are
 training one brain for communicating among connected sensors (agents), as they combine
 the neural networks of all agents in the training process.
 
+
+
+\begin{figure}
+  \centering
+  \fbox{\rule[-.5cm]{0.0cm}{4cm} 
+    \includegraphics{fig/occupy.pdf}
+    \rule[-.5cm]{0.5cm}{0.0cm}}
+  \caption{(a) A referential game example~\cite{??}. (b) Training procedure.}
+\end{figure}
+
 Roughly, to evolve a symbolic language, previous works force the
 agents to finish a set target through cooperation, which requires communication
 among the agents.
 These works can be classified into two categories based on the
 environment settings: \emph{referential games} and \emph{multi-agent
-  reinforcement learning system}. 
+reinforcement learning system} (MARL).
+%referential game
+In \emph{referential games}, agents are divided into \emph{sender} and
+\emph{receiver}, for speaking and listening, respectively. As the
+the referential game example shown in Figure~\ref{fig:rg}, one agent (Agent A) sends
+description of a target picture to another agent (Agent B), who will identify
+the target picture from a set of pictures~\cite{??}. However, in training, \note{xxxxxx}.
+%marl
+In \emph{MARL}, agents are placed in a virtual environment to cooperate in a
+continuous action space. For generating symbolic language, agents share the
+model parameters and/or environment information. Therefore, those agents can be
+taken as different sensors connected to a huge brain, not separate, individual
+brains. 
+
+Moreover, despite the individual agents issue, two more flaws existed in previous works.
+%intention
+First, intention is not considered in the cooperation among agents. Previous
+works always allocate each agent a role, either sender or receiver, for forced
+communication without considering their intention.
+%
+Second,
 
 
+In this paper, to achieve the naturally emergence of symbolic language among
+individual agents, we propose a novel Self-grounding-Introspection-Cooperation
+(SIM) model. There are four key difference between SIC model and previous
+works. First, to the best of our knowledge, SIC is the first work to generate
+symbolic language among \emph{individual} agents. Second, SIC is the first to
+achieve the naturally emergence of symbolic language. 
+ 
+
 \section{The Self-grounding-Introspection-Cooperation Model}
 
 
+
 \section{Experiments}
 
 \subsection{Methodology}