Skip to main content

Hypothesis Learning in Automated Experiments


Invention Reference Number

Process optimization


This disclosure describes technologies directed to model selection and deployment in automated experiments and fabrication processes. The disclosed technologies enable the ordering of model candidates based on their capability to describe relationships between measurements of a physical characteristic and an input parameter space. Each model candidate, associated with a prediction probability, undergoes a two-stage ranking process—pre-ranking to determine an initial reward and subsequent ranking to finalize the reward based on updated model parameters. This process employs active learning, significantly reducing the time and resources required for model selection and enhancing the efficiency of experiments or fabrications.


  • Efficiency improvement: Reduces the time required to identify the most accurate model by employing a structured active learning approach.
  • Reduced risk of damage: Minimizes the risk of damaging the sample material by limiting unnecessary experimentation.
  • Enhanced accuracy: Updates model parameters during the selection process, improving the accuracy of the chosen model for deployment.
  • Flexibility: Applicable to a wide range of experiments and fabrication processes, including those involving microscopes and spectroscopes.

Applications and Industries

  • Scientific research: Enhancing the selection of models in experiments involving complex physical characteristics measurements.
  • Industrial manufacturing: Optimizing fabrication processes by accurately selecting models for product design and development.
  • Material science: Applying in the synthesis and analysis of new materials with desired properties.
  • Medical research: Improving the efficiency of experiments in drug development and other biological research areas.

Related ORNL Technologies


To learn more about this technology, email or call 865-574-1051.