Machine learning for improved image-based wavefront sensing.

For large amounts of wavefront error, gradient-based optimization methods for image-based wavefront sensing are unlikely to converge when the starting guess for the wavefront differs greatly from the true wavefront. We use machine learning operating on a point-spread function to determine a good initial estimate of the wavefront. We show that our trained convolutional neural network provides good initial estimates in the presence of simulated detector noise and is more effective than using many random starting guesses for large amounts of wavefront error.