The speed and accuracy exceed that of humans. AI alone created 41 new materials in just 17 days.

"Nature" published two blockbuster studies on November 30: The latest artificial intelligence-driven platform GNoME (Graphic Network for Materials Exploration) can already discover and synthesize new inorganic compounds on its own, including the discovery of more than 2.2 million With a stable structure, he created 41 new materials on his own in 17 days, with speed and accuracy far exceeding that of humans.

This compound (Ba6Nb7O21) is one of the new materials calculated by GNoME. It contains barium (blue), niobium (white) and oxygen (green). Image source: Berkeley Lab Materials Program

Advances in technology have improved the ability of computer programs to identify new materials, but a major obstacle faced in the process is how learning algorithms adapt to results that are contrary to what they have learned. This is because new discoveries are essentially the ability to understand data in new, creative ways

The "Deep Thinking" team proposed a computing model this time that can improve the efficiency of material discovery through large-scale active learning. The program is trained using existing literature to generate a diverse set of candidate structures for potential compounds, which are then continuously refined through a series of learnings. GNoME discovered more than 2.2 million stable structures, improving the accuracy of structural stability predictions to over 80% and, in predicting composition, to 33% accuracy per 100 trials, compared with this figure in previous work. Only 1%.

In a second study, a UC Berkeley research team developed an automated laboratory (A-Lab) system. This A-Lab system is trained on existing scientific literature, combined with active learning, and can create up to five preliminary synthetic recipes for proposed compounds. It can then perform experiments with a robotic arm to synthesize the compound in powder form. If the yield of a recipe falls below 50%, A-Lab will adjust the recipe and continue experimenting until the goal is successfully achieved or all possible recipes are exhausted. After 17 days of continuous experimentation and 355 experiments, A-Lab successfully synthesized 41 of the 58 proposed compounds (71%). In comparison, human researchers would spend months guessing and experimenting

The training of AI demonstrated by the two research institutes combines the rapid development of computing power with existing literature. It proves that the use of learning algorithms to assist in the discovery and synthesis of inorganic compounds has extremely broad prospects. In the future, autonomous laboratories will be able to Discover new materials at the fastest speed with the least manpower.

(Source: Science and Technology Daily)

The above is the detailed content of The speed and accuracy exceed that of humans. AI alone created 41 new materials in just 17 days.. For more information, please follow other related articles on the PHP Chinese website!