Using graph neural networks in high-energy particle physics
Reference image and article address: [The next big thing: the use of graph neural networks to discover particles](https://news.fnal.gov/2020/09/the-next-big-thing- the-use-of-graph-neural-networks-to-discover-particles/)
The article describes the development of a set of graph neural network models by the Institute of High Energy Physics Fermilab in order to filter valuable images from the huge amount of particle collision images, which are deployed in the LHC lab to directly process the data generated during particle collisions.
In traditional digital images we use a dotted pixel structure, where each pixel is a combination of RGB red, green and blue, and using traditional convolutional neural networks, we can train the model well enough to be able to distinguish the characteristic properties in each image, however, in the field of molecules, scientists need to use X-rays to photograph the molecular structure of a chemical substance in order to be able to identify whether it is toxic or not. However, in the field of molecules, scientists need to use X-rays to photograph the molecular structure of a chemical substance in order to locate the physical properties of functional groups such as carbon rings and carboxyl groups. Studying 3D images of such X-rays is very different from the 2D images we see every day. In chemistry, we encounter many isomers, and if we use the 2D CNN idea, it is difficult to know the structure of the connections between atoms. relationship can be described by edge, so that we can quickly locate the features of molecular structure by GNNs model, unlike CNNs that use convolution kernels to scan the whole image for effective features in a fixed pixel region, but do not consider the pixel-to-pixel connection relationship.
Fermilab uses the GNNs model to quickly sift through the huge amount of images generated by high-energy particles colliding with each other every second, filtering out the useless images and thus verifying the proof of the existence of new particles more quickly.
Is it possible to use GNNs in the medical field, such as CT detection, Chest-XRay detection, disease prevention, most of the models currently seen used in the medical field are based on CNNs, is it possible to achieve more accurate and reliable judgments using GNNs, thus assisting doctors to quickly locate the disease condition of patients.