On April 16, the Aula Magna de La Salle Barcelona - Universitat Ramon Llull hosts the lecture by Andrey Ustyuzhanin, researcher at the Constructor University of Bremen and the Institute for Functional Intelligent Materials, National University of Singapore, and entitled 'Approaches to generating scientific hypotheses using graph neural networks and language models'.
With the rapid growth of scientific publications, innovative methods for opening scientific connections are emerging. The report combines two approaches - THiGER and HypoFinder. THiGER uses temporal graphs to analyze the dynamics of relationships between entities in the nutrition domain, while HypoFinder harnesses the power of large language models to automate the first stages of research and scientific literature analysis. Both approaches show the importance of integrating time graphs and language models to enhance the hypothesis generation process, offering new perspectives for scientific collaboration and discovery.
Andrey Ustyuzhanin is a researcher, data scientist and software developer. He has extensive experience working in different roles and with a variety of development and research methodologies, and has focused some of his research on finding applications of scientific approaches from the design of distributed and heterogeneous systems, to machine learning, to statistics in various areas such as particle physics, bioinformatics and robotics.