“Scientists from the University of Emory used a neural network, to determine the patterns of movement in dusty plasma – ionized gas with charged dust particles. The study is based on experiments in a vacuum chamber and a new SI method. According to scientists, they fully understand how and why this system works. “We have proven that AI is able to discover new physics. And what is important – this approach is universal […]”, – WRITE: Businessua.com.ua
Scientists from the University of Emory used a neural network, to determine the patterns of movement in dusty plasma – ionized gas with charged dust particles.
The study is based on experiments in a vacuum chamber and a new SI method. According to scientists, they fully understand how and why this system works.
“We have proven that AI is able to discover new physics. And importantly, this approach is universal and can be applied to other multi -part systems, ”said the co -author of the work Professor of Experimental Physics Justin Burton.
The neural network model, trained on three -dimensional trajectories of particles in dusty plasma, described asymmetrical (non -recyclable) forces with an accuracy of more than 99%. This made it possible to detect inaccuracies in common theoretical assumptions.
For example, it turned out that the charge of the particle did not grow strictly in proportion to its radius, as it was thought. Also, the dependence of the force of interaction between the particles on the distance in fact depends on the size of the particles – contrary to the established imagination.
Researchers compared the principle with the interaction of two boats on the water. When one goes in front, his waves can attract or push the other. In a dusty plasma, the leader attracts someone who is behind, but not vice versa, and this asymmetry now has an accurate explanation.
The team has developed a special tomographic system: the laser scans the “cut” of the space in the vacuum chamber, and the high -speed chamber fixes the position of the particles. The result is a 3D system motion model for several minutes.
The neural network simulates three types of forces: the viscosity of the environment, external forces and the interparticular interactions. And all this is on a regular desktop computer.
The authors hope that this approach will be the basis for the study of other complex systems: from paints and ink to biological fabrics. One of the leaders of the study, Professor Ilya Nemenman, plans to adapt a technique for analyzing the collective cell behavior.
Scientists emphasize that such studies still need expert people to properly build the architecture of the neural network, interpret the results and check their experiment
Recall that in July UCLA engineers with the help of AI created a new class of passive materials that can be structurally “programmed” to arbitrary control of light refraction.
The gun
Please wait …