É«É«À²

Kunal Ghosh (CEST group) explores data efficient machine learning methods in his doctoral studies. These methods enable applications in novel materials discovery such as organic electronics to build solar cells, as one example.

Ghosh has now received a one-year grant for doctoral students from the Finnish Cultural Foundation to continue his studies in this direction.

The research funded by this grant tackles the fundamental problem of how to compile good quality material datasets. Such datasets have become an important resource in materials science as they enable machine-learning based predictions of properties, for novel or improved materials. Ghosh is developing an active machine learning approach that explores a large materials space and iteratively compiles a target dataset for a specific materials design or optimization task. Datasets compiled with active learning are smaller than conventional datasets assembled by human intuition and facilitate more accurate and more targeted predictions. In his PhD, Ghosh is using his active learning algorithm to build up a dataset of organic molecules with optimal properties for application in organic electronics, e.g. organic light emitting or photovoltaic devices. In conclusion, machine learning models greatly benefit from better training datasets- and the outcome has the potential to impact the field of data-driven material science, where machine learning is now ubiquitously applied.