CLIENT | Parastoo Semnani (Machine Learning group) TU Berlin/BIFOLD

BRIEFING | graphical abstract, journal cover illustration and accompanying visualisations on a machine learning framework which optimises the search for new catalyst materials while accommodating small datasets and highly imbalanced data and simultaneously saving resources.

ADDITIONAL DETAILS about research content | particular elements and supports were combined into potential catalysts and tested in an oxidative methane coupling process (where methane was converted into natural gas using oxygen and the catalyst) The collected data was then put through the ML framework, where it was 1) classified: + (better performing) or – (worse performing), 2) cleaned: compensations made within data to balance + and – data, 3) used as a training set for the software.
4) further resampling was then used to balance out positive and negative, as well as fine-tuning and evaluation through examining specifics of feature importance in each element used.

AWARDS | 

• best poster at the 27th CSC 2024 University of Sherbrooke Quebec
• chosen for presentation at the ai grid women in ai science slam – ai data summit 2024
• ACS Editors’ Choice (highlighted by ACS for six months)
• chosen cover for Journal of Physical Chemistry (December 2024) acs publications