30 September 2019 to 4 October 2019
Europe/London timezone

Recursive neural tensor networks for jet classification

1 Oct 2019, 16:30
12m

Speaker

Mr Henry Day-Hall (University of Southampton)

Description

The prospect of using AI to utilise a greater fraction of the data available from colliders is very alluring,
particularly for events with limited statistics, such as Higgs decays.
On the topic of jet identification there are no shortage of attempts at such,
however, the `no free lunch' theorem is very central to use of AI.
Because of their boosted geometry, heavy highs decays can provide a particularly interesting
playground for a number of tools.
General techniques may solve many problems, but they will suffer more from noise and tendency to overfit.
A technique with the right geometry for the problem will
have a hypothesis space that better matches the correct solution.
I will discuss the use of Recursive Neural Tensor Networks for
jet classification or tagging, as these networks are an
excellent match for the shape of the problem.

Primary author

Mr Henry Day-Hall (University of Southampton)

Presentation materials