Uni-Logo

Seminar on Bayesian Optimization

The seminar language will be English (even if everyone is German-spoken, to practice presentation skills in English). First meeting: TBD.

Background

Bayesian optimization is a popular method for blackbox function optimization. Blackbox function are functions for which no assumptions are made, which means that neither the derivatives nor the smoothness are known. Furthermore, function evaluations might be noisy and are typically assumed to be expensive. Finally, the only knowledge about blackbox functions is the call signature which allows one to query the function for different input values (and observe the outcome).

These properties make Bayesian optimization an ideal method for hyperparameter optimization. In this seminar we will read papers on both the foundations of Bayesian optimization and recent research aiming to apply Bayesian optimization to state-of-the-art deep learning models.

Requirements

  • Machine Learning
  • Statistical Pattern Recognition
  • Automated Machine Learning would be good

Further information

For questions, please send an email to one of the organizers: Frank Hutter, Katharina Eggensperger, Matthias Feurer, Noor Awad, Arbër Zela