Researchers at Princeton University, have harnessed power from artificial intelligence (AI) to build new avenues to predict the properties of novel materials with improved accuracy and efficiency.

The method, outlined in the paper titled “LLM-Prop: Predicting Physical And Electronic Properties Of Crystalline Solids From Their Text Descriptions,” represents a shift in materials discovery. It relies on synthesising information from text descriptions, including intricate details such as the length and angles of bonds between atoms, as well as measurements of electronic and optical properties.

To train the model, researchers compiled a text benchmark containing descriptions of over 140,000 crystals, then leveraged an adapted version of Google Research’s large language model, T5, to predict the properties of various crystal structures, ranging from commonplace table salt to silicon semiconductors.

Up until now, traditional tools for crystal property prediction have predominantly utilised graph neural networks. However, these methods often fall short in capturing the nuances of crystal geometry and electronic properties.

“We have made tremendous advances in computer vision and natural language,” said Adji Bousso Dieng, an assistant professor of computer science at Princeton, “but we are not very advanced yet when it comes to dealing with graphs (in AI). 

Recommended

• Strathclyde Uni Project Looks at Ethical Risks of Gen AI in Research
• Edinburgh Uni Leading Charge for EV Battery Recycling Tech
• UK Looks Closer to Home for Green Tech Raw Materials

“So, I wanted to move from the graph to actually translating it to a domain where we have great tools already. If we have text, then we can leverage all these powerful (large language models) on that text.”

Craig Arnold, professor of mechanical and aerospace engineering and vice dean for innovation, emphasises the transformative potential of the language model-based approach.

He remarks, “It’s really about, how do I access all of this knowledge that humanity has developed, and how do I process that knowledge to move forward? It’s characteristically different than our current approaches, and I think that’s what gives it a lot of power.”