NLP for materials design
Question Answering Method to Extract Information from Materials Science Literature
Project Lead: Matilda Sipilä
Scientific text is a valuable data source for materials science, and development of large language models allows us to use new machine learning tools like question answering (QA). We developed and tested QA method for extracting bandgap values of perovskites from literature using BERT models. Our work demonstrates that QA is versatile, scalable, and user-friendly method to extract information from scientific text.
Partner(s): Prof. Filip Ginter and Dr. Sampo Pyysalo, TurkuNLP, Department of Computing, University of Turku.
Information Extraction from Scientific Tables in the Literature
Project Lead: Tomasz Galica
Scientific publications in material science are rich in tables that contain valuable data about the physical properties of materials. In this project, we leverage NLP methodologies to extract and analyze the underused tabular datasets in material science research.
Partner(s): Prof. Filip Ginter, TurkuNLP, Department of Computing, University of Turku
Materials design of inorganic crystals with 3D transformers
Project Lead: Tomasz Galica
Visualizing and mapping the materials space is a challenging task due to both the large volume and complexity of material data. We are exploring novel, non-atomistic visualization algorithms to better analyze the relationships between compounds and materials properties.
Partner(s): Prof. Filip Ginter, TurkuNLP, Department of Computing, University of Turku
MaterialsQA web application
Project Lead: Henri Haapanen
Substitution of atomic species on different crystal lattice sites can be used to design new materials and tune functional properties for devices. We borrow Transformer algorithms from NLP to learn the chemistry of inorganic materials and predict compatible element substitutions for materials design.
Partner(s): Prof. Filip Ginter, TurkuNLP, Department of Computing, University of Turku
Materials design of inorganic crystals with 3D transformers
Project Lead: Christer Söderholm
QA models allow non-expert users to extract information about materials properties from scientific publications with simple questions. We are developing a QA materials web application, based on our trained AI models, to bring the power of NLP to colleagues and wider audiences.
Partner(s): Prof. Filip Ginter, TurkuNLP, Department of Computing, University of Turku