Intelligent Experimentation

Scientific experiments are necessary to acquire new knowledge in natural and engineering sciences. Typically, experiments are designed in a step-wise approach, including design, execution, data processing, evaluation, and interpretation. Iterations for improving either of these steps are possible but typically cumbersome. To increase the efficiency of computer-controlled automation, exploiting an interactive control of all installed sensors and detectors offers the opportunity to optimize the general result by decreasing process times. Rapid evaluation methods such as machine learning techniques, e.g., deep learning models, enable direct analysis and monitoring of acquired data.

For this reason, we propose integrating an online data analysis system into scientific experiments, which can, as a result, be designed and evaluated on the fly and, therefore, aid in improving the experiment procedure over time. By implication, this should in addition allow the experiment to influence and optimize the ML design.

Central Research Questions of Intelligent Experimentation are:

  • How does integration of online data analysis into the experimental setting look like?
  • What can we extract from the experimental data during an experiment (on the fly) and how can we use this information to design the experiment?
  • Which parameters are best suited for the experiment?
  • When to stop the current experiment?