| 1 |
< |
The TaNC algorithm has been with the goal of maximizing the discriminatory |
| 2 |
< |
information available to the algorithm, and can be broken down into four |
| 3 |
< |
steps: |
| 4 |
< |
|
| 5 |
< |
\begin{itemize} |
| 6 |
< |
|
| 7 |
< |
\item Find tau--candidate seeds by applying a jet-finding |
| 8 |
< |
(iterativeCone5PFJet) algorithm. Within the jet, select objects |
| 9 |
< |
kinematically consistent (according to the defintion of the ``shrinking |
| 10 |
< |
cone'' tau algorithm in ~\ref{PFT08001}) with a tau decay as ``signal'' |
| 11 |
< |
objects. |
| 12 |
< |
|
| 13 |
< |
\item Merge photon pairs in the signal object collection into candidate |
| 14 |
< |
neutral pions. |
| 15 |
< |
|
| 16 |
< |
\item Remove objects that are consistent with underlying event contamination. |
| 17 |
< |
|
| 18 |
< |
\item Determine the decay mode using the multiplicity of charged and neutral |
| 19 |
< |
objects. |
| 20 |
< |
|
| 21 |
< |
\item Apply a discriminant \emph{specific to the reconstructed decay mode} to |
| 22 |
< |
the tau candidate. |
| 23 |
< |
|
| 24 |
< |
\end{itemize} |
| 25 |
< |
|
| 26 |
< |
|
| 1 |
> |
The Tau Neural Classifier algorithm reconstructs the decay mode of the |
| 2 |
> |
tau--candidate and then feeds the tau--candidate to a discriminator associated |
| 3 |
> |
to the decay mode to make the classification decision. Each discriminator |
| 4 |
> |
therefore maps to a reconstructed decay mode in a one-to-one fashion. To |
| 5 |
> |
optimize the discrimination for each of the different decay modes, the TaNC uses |
| 6 |
> |
an ensemble of neural nets. Each neural net corresponds to one of the dominant |
| 7 |
> |
hadronic decay modes of the tau lepton. These selected hadronic decays |
| 8 |
> |
constitute 95\% of all hadronic tau decays. Tau--candidates with reconstructed |
| 9 |
> |
decay modes not in the set of dominant hadronic modes are immediately tagged as |
| 10 |
> |
background. |
| 11 |
|
|
