Name Surname
Leveraging Large Language Models to for Automated Data Explanations in Petrology
Main Supervisor: : Blaž Zupan (University of Ljubljana)
Co-Supervisors: Maurizio Petrelli (University of Perugia), Efstratios Gavves (University of Amsterdam)
Objectives
This project aims to develop models that can create explanations or summarize various data modalities in petrology that would be useful for experts. The modalities will include microscope images, single-spot chemical compositions, chemical imaging, and possibly others. The models will be trained on the literature, public databases (such as Virtual Microscope and GEOROC), and data obtained by REALISE project partners. We will also attempt to build explanation models combining multiple data modalities, thus performing data fusion. The planned explanation models will be based on data embedding techniques and generative AI, but given the rate of development of the field, the specifics will be defined during the project. The resulting models will not focus on a particular data source; instead, they will be targeted towards the field of petrology in general, with special focus to the exploitation of Critical Raw Materials.
Epected Results
- Models that can perform explanations in the field of petrology for various data modalities;
- The integration of the models performing explanations into a data analysis tool based on visual programming.