Fig. 3
From: Multiparameter optimisation of a magneto-optical trap using deep learning
Experimental run and additional solutions. a A typical experimental run consists of four stages to generate an atomic cloud. Initially 500 ms worth of loading time is employed to statically load atoms into a trap. After this the compression sequence generated from a parameter set is performed over 20 ms. Following this a 1 ms preparation stage utilises PGC cooling and optical pumping schemes to the right magnetic level. Finally a measurement stage provides experimental feedback for the cost function via measurement of the probe absorption. The entire experimental run is repeated at a rate of approximately 2 Hz. The outcome of the measurement stage is monitored to determine when an equilibrium has been reached. b Additional sub-optimal parameter sets found by the SANN. The corresponding OD relative to the maximum OD achieved is shown above each solution
