Motivation
<l2st>
AD is super important...
good AD = complicated models -> Many Parameters
Evaluation dependent on very few datapoints->Optimization impossible
</l2st>
Open Challenges
<l2st>
Evaluate without testing data
Formalisation of existing ideas
Numerical assessment of them
</l2st>
Contribution
<l2st>
Suggest new methods for AE
Compare methods on many datasets
Seperate into parameter and hyperparameter optimisation
</l2st>