Research Press Release
Nature
November 30, 2023
Two papers demonstrate the potential for using artificial intelligence-powered platforms to increase the speed and precision of discovering and synthesizing new inorganic compounds.NaturePublished in
Recent advances in technology have improved the power of computer programs to identify many new materials. However, because the discovery of new materials fundamentally requires the ability to interpret data in new and creative ways, one ability of learning algorithms is the ability to adapt to results that contradict already learned knowledge. is an impediment to the process of discovering new materials.
Now, Ekin Cubuk and colleagues present a computer model that increases the efficiency of new materials discovery through large-scale active learning. The program was trained using existing literature to generate a variety of structural candidates that could lead to new compounds, and its accuracy was improved through rounds of active learning. Using these models (which Cubuk et al. call “graph networks for materials exploration”), more than 2.2 million stable structures have been discovered, with an accuracy of 80% in predicting structure stability. The prediction of composition improved by 33% per 100 trials (compared to 1% in previous studies).
Meanwhile, Gerbrand Ceder and colleagues have developed the Autonomous Laboratory (A-Lab) system. A-Lab is trained using existing scientific literature and, using active learning, can create up to five initial synthesis recipes for a given compound, and then uses a robotic arm to create Experiments can be performed to synthesize powdered compounds. If the synthesis yield using one recipe is less than 50%, A-Lab will modify the recipe and continue the experiment, and if the goal is achieved or all possible recipes have been exhausted. Terminate the experiment if Over 17 days of continuous experimentation, A-Lab performed 355 experiments and succeeded in synthesizing 41 (71%) of the 58 compounds instructed. Ceder et al. also showed that a small modification to the decision-making algorithm could increase the success rate to 74%, and a similar improvement in the computational method could further increase the success rate to 78%. Ta.
These two papers show that improvements in the data processing power of computers and training on existing literature have led to promising advances in the use of learning algorithms to aid in the discovery and synthesis of inorganic compounds. It shows.
doi:10.1038/s41586-023-06734-w
“Highlights from Nature Magazines” is a translation of a release prepared for the press by the Nature Public Relations Department. If you require more accurate and detailed information, please be sure to refer to the original paper.
Return to “Notable Highlights” article list
