Please read these instructions before posting any event on Fermilab Indico

Indico will be unavailable on Wed., Dec 18th from 7 - 7:30am due to server maintenance.

22–28 Jul 2018
Kellogg Hotel and Conference Center
EST timezone

Do not measure correlated observables, but train an artificial intelligence to predict them

26 Jul 2018, 08:30
20m
106 (Kellogg Hotel and Conference Center)

106

Kellogg Hotel and Conference Center

219 S Harrison Rd, East Lansing, MI 48824
Hadron Structure Hadron Structure

Speaker

Dr Boram Yoon (Los Alamos National Laboratory)

Description

In lattice QCD calculations, many different observables are measured on a gauge field, and their statistical fluctuations are correlated. By exploiting the correlation, one observable can be reconstructed from other observables, without expensive direct calculation. This idea is applied to two nucleon matrix element calculations using machine learning technique. (1) The calculations of nucleon charges and form-factors need observables at multiple separations of nucleon source and sink in Euclidean time ($t_\text{sep}$) to remove excited state contamination. We trained a boosted decision tree regression machine learning algorithm to predict observables at $t_\text{sep} = 8a$ and $10a$ from the observables at $ t_\text{sep} = 12a$ on a $a=0.09$ fm lattice. (2) In the Schwinger source method for the quark chromo-electric dipole moment (cEDM), nucleon matrix elements are calculated from the quark propagators including CP-violating operators. We trained a machine to predict two-point correlation functions of the cEDM- and $\gamma_5$-inserted quark propagators from those of normal quark propagators without CP-violating operators.

Primary author

Dr Boram Yoon (Los Alamos National Laboratory)

Presentation materials