This technology is a software that translates naturally occurring, human-described action into a machine-readable format for process analysis, optimization, and discovery using machine learning.
Many critical processes in manufacturing and R&D are guided by human expertise and described through natural language, such as a lab protocol or recipe. While computational tools exist to optimize well-defined, repetitive systems, they cannot interpret these unstructured, "naturally occurring" actions to systematically find efficiencies or discover new pathways. This creates a barrier to applying powerful machine learning techniques for broad process optimization, leaving improvements to rely on individual trial-and-error and "intuition."
This technology is a software that interprets a naturally occurring sequence of actions (e.g., a chemical synthesis recorded in an electronic lab notebook) and translates it into a directed acyclic graph (DAG) that can be analyzed using machine learning. The DAG-based representation is converted to a numerical vector suitable for analysis. Machine learning models can be applied to a database of these process DAGs and learn associations between graph features and measured outcomes to identify causal patterns, reduce parameter space, and recommend simplified or alternative sequences. These insights allow users to optimize existing procedures, dynamically adapt processes to changing circumstances, or discover novel pathways to achieve a desired outcome.
IR CU20144
Licensing Contact: Greg Maskel
