conclusion.tex

% !TEX root = recipeUnderstanding.tex


\section{Conclusions}
\vspace{-2mm}
%Discuss which recipes worked and why. Discuss the importance of semantic representation, scaling features and multi-modality.
In this paper, we tried to capture the underlying structure of human communication by jointly considering visual and language cues. We experimentally validated that given a large-video collection having subtitles, it is possible to discover activities without any supervision over activities or objects. Experimental evaluation also suggested the available noisy and incomplete information is powerful enough to not only discover activities but also describe them. We also think that the resulting discovered knowledge can be effectively used in many domains like multimedia interfaces and robot knowledge bases \cite{robobrain}. 
\vspace{-2mm}
\section{Acknowledgements}
\vspace{-2mm}
We acknowledge the support of ONR award N00014-13-1-0761 and ONR award N000141110389.
\vspace{-2mm}