麻豆淫院

Many properties of the world's most advanced materials are beyond the reach of quantitative modeling. Understanding them also requires a human expert's reasoning and intuition, which can't be replicated by even the most powerful artificial intelligence, mixed with fortuitous accident, according to Eun-Ah Kim, the Hans A. Bethe Professor of physics in the College of Arts and Sciences.

Kim and collaborators have developed a machine-learning model that encapsulates and quantifies the valuable intuition of human experts in the quest to discover new quantum materials. The model, Materials Expert-Artificial Intelligence (ME-AI), "bottles" this intuition into descriptors that predict the functional properties of a material. The team used the method to solve a quantum materials problem.

"We are charting a new paradigm where we transfer experts' knowledge, especially their intuition and insight, by letting an expert curate data and decide on the fundamental features of the model," said Kim, director of the Cornell-led National Science Foundation AI-Materials Institute. "Then the machine learns from the data to think the way the experts think."

Kim is the corresponding author of "Materials Expert-Artificial Intelligence for Materials Discovery" in Communications Materials.

The study, a collaboration with Leslie Schoop, associate professor of chemistry at Princeton University, formed a foundation for AI-MI's vision toward discovering next-generation materials through targeted search, as opposed to serendipitous discovery. AI is a necessary component of this data-heavy quest, but it needs to be incorporated strategically, Kim said.

With the help of AI, vast volumes of information accumulated through human experiences can be combed to uncover qualities that predict desired properties, but indiscriminate collection of sources that are not guided by an expert's intuition can be misleading, according to the researchers.

To do this work, the researchers identified a specific problem. Trying to find which of a group of 879 materials shared a certain, desirable characteristic, they trained a machine learning model using data curated and labeled by Schoop and her research group.

In the results, the ME-AI model reproduced the human expert intuition and expanded upon it. In addition, the ME-AI model demonstrated an exciting generalization, predicting similar materials among a different set of compounds.

"What we found is that this framework essentially reproduced Leslie's insight, but it gave us more to chew on," Kim said. "It proved that when the researcher's approach to the data was really actually impactful, that same criteria can be reproduced by a machine."

In fact, when the model came up with some insight it hadn't been explicitly asked to produce, Schoop recognized her own thought process at work, saying, "Oh, that makes a lot of sense."

"The access we have to the human brain is very limited," Kim said. "When a human has a gut feeling, it happens too quickly for them to spell it out. They know it's right, but they wouldn't necessarily articulate their process."

In contrast, a machine is very good at explaining how it's reached a conclusion, she said. The researchers' vision for ME-AI is to give the machine insight into the human expert's process so that process becomes apparent in the conclusions.

The study provides a model for future collaborations at AI-MI, which matches materials scientists who study quantum physics and chemistry with computer scientists who have expertise in machine learning.

"AI-MI is at the frontier of using AI for discovery and learning about materials," Kim said. "Good data curation is everything if you want to make progress toward scientific discovery."