Artificial neural networks for laser frequency stabilization

In order to stabilize a laser's emission frequency, absolute references such as molecular absorption lines are widely used. To automate the stabilization process, the desired absorption line needs to be identified reliably from a spectrum by a computer. We present an artificial neural network solving this task using the iodine spectrum as an example. The neural network is trained using only simulated data and subsequently tested using measured data. We show that this approach is robust against large variations of operating and environmental conditions.