| 1 |
friis |
1.1 |
\documentclass{article}
|
| 2 |
|
|
|
| 3 |
|
|
\title{Tau Neural Classifier Note Outline}
|
| 4 |
|
|
\author{Evan K. Friis}
|
| 5 |
|
|
|
| 6 |
|
|
\begin{document}
|
| 7 |
|
|
|
| 8 |
|
|
\maketitle
|
| 9 |
|
|
\tableofcontents
|
| 10 |
|
|
|
| 11 |
|
|
\section{Introduction}
|
| 12 |
|
|
Standard tau physics intro.
|
| 13 |
|
|
|
| 14 |
|
|
\subsection{Traditional tau ID}
|
| 15 |
|
|
Reference PFT-08-001, QCD difficulty, discuss shrinking cones.
|
| 16 |
|
|
|
| 17 |
|
|
\subsection{TaNC motivation}
|
| 18 |
|
|
Different topologies and their corresponding resonances.
|
| 19 |
|
|
Large parameter space - use neural nets!
|
| 20 |
|
|
|
| 21 |
|
|
\section{Decay Mode Classification}
|
| 22 |
|
|
First step of the algorithm is correctly identifying the decay mode.
|
| 23 |
|
|
\subsection{Kinematic Envelope}
|
| 24 |
|
|
Shrinking cone description.
|
| 25 |
|
|
\subsection{Quality Cuts}
|
| 26 |
|
|
\subsection{Photon Merging}
|
| 27 |
|
|
\subsection{Performance}
|
| 28 |
|
|
Accuracy matrix (Reco DM vs. True DM); Invariant mass resolution with and without mass hypothesis. Perhaps show
|
| 29 |
|
|
improvement of seperation between taus and QCD with and without DM selection.
|
| 30 |
|
|
|
| 31 |
|
|
\section{Neural network classification}
|
| 32 |
|
|
Intro, neural nets are...
|
| 33 |
|
|
\subsection{Network topologies}
|
| 34 |
|
|
Choice of variables. Variable distributions. Optimization of variables. (Mike Squires?)
|
| 35 |
|
|
\subsection{Individual mode performance}
|
| 36 |
|
|
\subsection{Bayesian Neural Networks}
|
| 37 |
|
|
On results w/ neurobayes if time?
|
| 38 |
|
|
\subsection{Performance Space Mapping}
|
| 39 |
|
|
On Christian's technique to map the 5-d cuts to a single number
|
| 40 |
|
|
|
| 41 |
|
|
\section{Performance}
|
| 42 |
|
|
\subsection{Performance Points}
|
| 43 |
|
|
Discussiong of how cuts are selected.
|
| 44 |
|
|
\subsection{Results}
|
| 45 |
|
|
|
| 46 |
|
|
\end{document}
|
