Machine learning optimal control pulses in an optical quantum memory experiment

Abstract: Efficient optical quantum memories are a milestone required for several quantum technologies including repeater-based quantum key distribution and on-demand multi-photon generation. We present an efficiency optimization of an optical electromagnetically induced transparency (EIT) memory experiment in a warm cesium vapor using a genetic algorithm and analyze the resulting waveforms. The control pulse is represented either as a Gaussian or free-form pulse, and the results from the optimization are compared. We see an improvement factor of 3(7)\% when using optimized free-form pulses. By limiting the allowed pulse energy in a solution, we show an energy-based optimization giving a 30% reduction in energy, with minimal efficiency loss.