The neural network shortcut
By training on available data, neural networks learn algorithms that approximate the solution to a problem. It is not always clear how they do this. Now, two young Basel-based physicists are able to derive mathematical expressions which allow them to calculate the best solution without having to train a neural network. The results of their research not only provide insight into the workings of learning algorithms, but they could also be used to detect phase transitions that are unknown in future physical systems.
The neural networks work on the same principle as the brain. These computer algorithms can learn to solve problems by repeated training. They can, for instance, distinguish objects and process spoken language.
Since several years, physicists are also trying to use neural network to detect phase changes. We are all familiar with phase transitions, such as when water turns to ice. But they can also be more complex, occurring between phases of magnetic materials and quantum systems.
Source:
https://techxplore.com/news/2022-09-shortcut-neural-networks.html
Leave a Reply