\documentclass[10pt]{article}
\usepackage{fullpage}
\usepackage{setspace}
\usepackage{parskip}
\usepackage{titlesec}
\usepackage[section]{placeins}
\usepackage{xcolor}
\usepackage{breakcites}
\usepackage{lineno}
\usepackage{hyphenat}
\PassOptionsToPackage{hyphens}{url}
\usepackage[colorlinks = true,
linkcolor = blue,
urlcolor = blue,
citecolor = blue,
anchorcolor = blue]{hyperref}
\usepackage{etoolbox}
\makeatletter
\patchcmd\@combinedblfloats{\box\@outputbox}{\unvbox\@outputbox}{}{%
\errmessage{\noexpand\@combinedblfloats could not be patched}%
}%
\makeatother
\usepackage[round]{natbib}
\let\cite\citep
\renewenvironment{abstract}
{{\bfseries\noindent{\abstractname}\par\nobreak}\footnotesize}
{\bigskip}
\titlespacing{\section}{0pt}{*3}{*1}
\titlespacing{\subsection}{0pt}{*2}{*0.5}
\titlespacing{\subsubsection}{0pt}{*1.5}{0pt}
\usepackage{authblk}
\usepackage{graphicx}
\usepackage[space]{grffile}
\usepackage{latexsym}
\usepackage{textcomp}
\usepackage{longtable}
\usepackage{tabulary}
\usepackage{booktabs,array,multirow}
\usepackage{amsfonts,amsmath,amssymb}
\providecommand\citet{\cite}
\providecommand\citep{\cite}
\providecommand\citealt{\cite}
% You can conditionalize code for latexml or normal latex using this.
\newif\iflatexml\latexmlfalse
\AtBeginDocument{\DeclareGraphicsExtensions{.pdf,.PDF,.eps,.EPS,.png,.PNG,.tif,.TIF,.jpg,.JPG,.jpeg,.JPEG}}
\usepackage[utf8]{inputenc}
\usepackage[english]{babel}
\begin{document}
\title{ECE498LV Project Proposal: Bottlenecking as a Metric for Effective Collaborative Networks}
\author[1]{Naren Dasan}%
\author[1]{Tanishq Dubey}%
\affil[1]{University of Illinois at Urbanaâ€“Champaign}%
\vspace{-1em}
\date{\today}
\begingroup
\let\center\flushleft
\let\endcenter\endflushleft
\maketitle
\endgroup
\sloppy
We are looking determine if bottlenecked flow is a property correlated
with productive networks and to do so we will analyze the flow of
artificial neural networks and naturally occurring collaborative
networks. Some learning theorists believe the the generalization
properties of neural networks are due to a phenomenon known as
information bottlenecking~\citep{Tishby_2015}~\citep{shwartz2017opening} and
machine learning scientists have found that important activations in
artificial neural networks are concentrated in a few
nodes~\citep{radford2017learning}. We want to explore these properties of neural
networks and then compare them to what occurs in naturally occurring
networks focusing mostly on collaboration to see if bottlenecking is a
property that could be correlated with productivity. We are going to
start by analyzing the flow in neural networks to verify these ideas of
bottlenecked flow. Then we intend to run a similar analysis on analogous
networks such as hierarchical networks and other types of networks
describing collaborative situations. Finally we will evaluate random
networks for the same properties and also determine their effectiveness
in machine learning tasks as a baseline for the two other classes of
networks.~ We hope to see that effective networks for a particular task
exhibit the property of highly bottlenecked flow and that networks that
don't exibit this property are not as effective.~
We intend to accomplish this by observing the dominant paths of an ANN
though tools such as TensorBoard and techniques described in the
literature~\citep{zeiler2014visualizing}~\citep{simonyan2013deep} as well as basic flow
analysis. With this, we can then observe the most prominent edges and
find the bottlenecks and the semantically important nodes that have
developed within the network. We will then attempt to represent
naturally occuring networks and random networks in a similar fashion to
ANNs to run the same analysis. Some of the ANN networks we are going to
examine include AlexNet, VGG, and GoogleNet as well as others, datasets
we are examining include the metagenome of
stromatolites~\citep{Mobberley2015}, the~\textbf{polymath network, moth
olfactory connectome} and some of the networks described
in~\citealt{corominas2013origins}. We are also determining if citation networks or
networks describing academic families fit into the class of networks we
are studying.~
\par\null
We think bottlenecking as a property of a network has interesting
repercussions. It may be that because flow becomes so concentrated,
network flexibility is a natural response. Showing that ANNs and
effective naturally developed collaborative networks both exhibit this
property might give network engineers a new metric to evaluate the
potential performance of a system.
\par\null
We also need to resolve the distinction between information flow
bottlenecking and bottlenecked flow. Look for DAG like structures
\href{http://groups.csail.mit.edu/pag/pubs/secret-max-flow-pldi2008.pdf}{http://groups.csail.mit.edu/pag/pubs/secret-max-flow-pldi2008.pd}\href{http://groups.csail.mit.edu/pag/pubs/secret-max-flow-pldi2008.pdf}{f}
\url{https://www.nature.com/articles/s41467-017-01916-3}
\url{http://gow.epsrc.ac.uk/NGBOViewGrant.aspx?GrantRef=EP/H016015/1}
\par\null
\selectlanguage{english}
\FloatBarrier
\bibliographystyle{plainnat}
\bibliography{bibliography/converted_to_latex.bib%
}
\end{document}