We are a versatile group in artificial intelligence research. Most of our work falls under the umbrella of new methods for (semi-)automating the scientific design and analysis of algorithms, and their applications. Here are some pointers to ongoing projects:

  • AutoML: Automatic Machine Learning
    • AutoWEKA: a system offering off-the-shelf high-performance supervised machine learning
    • HPOlib: hyperparameter optimization library, for the systematic evaluation of different optimizers, prominently including deep learning benchmarks
    • Automated Hyperparameter Importance Analysis: using the same models as Bayesian optimization to assess parameter importance
  • Algorithm Configuration: systems for finding high-performance parameter settings of a given algorithm
    • ParamILS: iterated local search in parameter configuration space.
    • SMAC: sequential optimization in parameter configuration using response surface methods.
  • Algorithm Selection: SATzilla: a system for selecting the right SAT algorithm for new problem instances.
  • Empirical Algorithmics:Empirical Performance Models: statistical models capturing how algorithm performancedepends on instance features and algorithm parameters
  • Configurable SAT Solver competition (CSSC): this competition we organize rewards SAT solvers that perform best after being configured automatically for particular application domains.