Machine learning models for energetic materials properties using multi-task learning

The emergence of data science and artificial intelligence has created a new paradigm for how science is being conducted. In materials science, an abundance of resources is invested into developing infrastructure to storing scientific data that is easily accessible for other researchers. However, in many fields such as energetic materials, there is a lack of organized data structured in a way that makes application of these advanced techniques straightforward. We have identified rich sources of experimental and calculated data specifically focused on energetic materials and have collected this data into an electronic format that is tailored for efficient querying, filtering, and extracting. This allows us to apply machine learning models in a simple manner but also serves as a powerful resource to the general researcher in our field. We present models leveraging the multi-task learning approach to predict multiple energetic materials properties simultaneously.