AI Speeds Up Discovery of New Materials

Russian researchers have made a significant breakthrough in the development of new materials. They utilized Artificial Intelligence (AI) to accelerate the process of identifying and analyzing the properties of prospective compounds. A team from the AIRI Institute of Artificial Intelligence, with support from Sber, Skoltech, and Tomsk Polytechnic University, created an AI system based on graph neural networks (GNNs). This system helps find effective combinations of elements for creating promising new compounds based on boron and wolfram. "The trained model allowed us to analyze all the data in just a few days and select the most promising ones for experimental verification," says Roman Eremin, leading researcher at AIRI. Traditional methods of calculating the properties of molecules, which are quantum-chemical, require significant computational resources and time. Especially if the structure of the compound is complex -- with each atom and electron calculation becoming more difficult. This can take months or even years. AI helps to bypass this barrier. This allows you to speed up the process and not waste hundreds of thousands of configurations in vain. "We have consistently included in the training only those structures where the model has made the most mistakes. This reduced the combinatorial complexity of the task," explains Eremin. The created AI model is not limited only to boron compounds. It can be used to search for new materials in other chemical systems -- from medical compounds to space propulsion. A neural network (also an artificial neural network, INS, or simply a neural network) is a mathematical model, as well as its software or hardware implementation, built on the principle of the organization of biological neural networks -- the networks of nerve cells of a living organism.