Motivation AD is super important... good AD = complicated models -> Many Parameters Evaluation dependent on very few datapoints->Optimization impossible Open Challenges Evaluate without testing data Formalisation of existing ideas Numerical assessment of them Contribution Suggest new methods for AE Compare methods on many datasets Seperate into parameter and hyperparameter optimisation