Dimitris N. Chorafas Foundation Award - 2023 - Kevin Jablonka
Toward data-driven materials design: From atoms to pilot plants
Thesis director: Prof. B. Smit
For creating an ecosystem for machine learning in chemistry and chemical engineering that allowed him to address problems beyond our current understanding
To address many of the most pressing societal needs we need new materials.
However, there are too many materials on could potentially test. In combination with the fact that for many phenomena we still only have an incomplete understanding, this makes brute-force screenings using experimental or computational tools completely unfeasible.
A promising alternative is materials design using data-driven techniques. In this case, we build efficient predictive models by learning the relationship between structures and properties from data.
Such a data-intensive approach, however, has unique challenges. First, it relies on suitable data. Second, appropriate tooling is needed to ensure that research findings can be easily compared and reused. A third challenge is that crystal structures often must be first converted into suitable inputs for machine learning models (so-called featurization).
The thesis addresses all these challenges by introducing an “ecosystem” of tools for digital reticular chemistry, including electronic lab notebooks, datasets and featurization techniques. In combination with novel search approaches, applied these techniques from the atom to the pilot plant scale.
The progress thus far indicates that machine learning might have a larger impact on chemistry than in many other domains, such as computer vision.