For years, physicists have been making major advances and breakthroughs in the field using their minds as their primary tools. But what if artificial intelligence could help with these discoveries?
Last month, researchers at Duke University demonstrated that incorporating known physics into machine learning algorithms could result in new levels of discoveries into material properties, according to a press release by the institution. They undertook a first-of-its-kind project where they constructed a machine-learning algorithm to deduce the properties of a class of engineered materials known as metamaterials and to determine how they interact with electromagnetic fields.
Predicting metamaterial properties
The results proved extraordinary. The new algorithm accurately predicted the metamaterial’s properties more efficiently than previous methods while also providing new insights.
“By incorporating known physics directly into the machine learning, the algorithm can find solutions with less training data and in less time,” said Willie Padilla, professor of electrical and computer engineering at Duke. “While this study was mainly a demonstration showing that the approach could recreate known solutions, it also revealed some insights into the inner workings of non-metallic metamaterials that nobody knew before.”
In their new work, the researchers focused on making discoveries that were accurate and made sense.
“Neural networks try to find patterns in the data, but sometimes the patterns they find don’t obey the laws of physics, making the model it creates unreliable,” said Jordan Malof, assistant research professor of electrical and computer engineering at Duke. “By forcing the neural network to obey the laws of physics, we prevented it from finding relationships that may fit the data but aren’t actually true.”
They did that by imposing upon the neural network a physics called a Lorentz model. This is a set of equations that describe how the intrinsic properties of a material resonate with an electromagnetic field. This, however, was no easy feat to achieve.
“When you make a neural network more interpretable, which is in some sense what we’ve done here, it can be more challenging to fine tune,” said Omar Khatib, a postdoctoral researcher working in Padilla’s laboratory. “We definitely had a difficult time optimizing the training to learn the patterns.”
A significantly more efficient model
The researchers were pleasantly surprised to find that this model worked more efficiently than previous neural networks the group had created for the same tasks by dramatically reducing the number of parameters needed for the model to determine the metamaterial properties. The new model could even make discoveries all on its own.
Now, the researchers are getting ready to use their approach on unchartered territory.
“Now that we’ve demonstrated that this can be done, we want to apply this approach to systems where the physics is unknown,” Padilla said.
“Lots of people are using neural networks to predict material properties, but getting enough training data from simulations is a giant pain,” Malof added. “This work also shows a path toward creating models that don’t need as much data, which is useful across the board.”
The study is published in the journal Advanced Optical Materials.