Abstract: ‘Experiments whose data are processed using a computational model are often plagued by uncertainties in the experiment which prohibit accurate outcomes of the model. As a result, experimental physicists spend their time in labs painstakingly calibrating their setups to minimize any form of uncertainty in the model. Occasionally they are so fed up with this that they try to include the parameter into their models instead, but this often requires some theoretical work to find a suitable optimization approach. In this presentation we will see that automatic differentiation can provide a general means of optimization: no additional theory required. By means of example, we show this works for the case of the tilt angle in reflection ptychography.’
a joint institute of 
and 
and