Scientists create artificial neural networks that detect symmetry and patterns

A probe squad astatine Lehigh University, funded by the U.S. National Science Foundation, developed and efficaciously taught an artificial neural web to consciousness symmetry and structural similarities successful materials and to make similarity projections. The researchers published their findings successful the diary npj Computational Materials.

The squad developed an artificial neural network and utilized instrumentality learning to bid the neural web to spot symmetry and observe patterns and trends. In the archetypal effort of its kind, the researchers utilized this innovation to hunt a database of much than 25,000 images and successfully classified akin materials. The web could alteration materials probe by analyzing tremendous amounts of accusation and information from experiments to observe and decode patterns successful multidimensional data.

"If you bid a neural network, the effect is simply a vector, oregon a acceptable of numbers that is simply a compact descriptor of the features," said Joshua Agar, a co-author and instrumentality learning idiosyncratic astatine Lehigh University. "These features assistance classify things truthful that immoderate similarity is learned. What's produced is inactive alternatively ample successful space, though, due to the fact that you mightiness person 512 oregon much antithetic features. So, past you privation to compress it into a abstraction that a quality tin comprehend specified arsenic 2D oregon 3D—or possibly 4D."

The artificial neural network could assistance scientists and researchers larn much astir the multidimensional operation of materials and the complexities of structure-property dynamics. Artificial neural networks could analyse images and information from failed experiments and let materials researchers to find structural similarities, patterns and trends successful research data. With improved data absorption and accessibility, that could uncover undetected trends and patterns, summation experimentation ratio and accelerate research.

---

---

More information: Nguyen, T.N.M et al, Symmetry-aware recursive representation similarity exploration for materials microscopy. npj Comput Mater (2021). doi.org/10.1038/s41524-021-00637-y