A representation of the variety of materials under investigation for their potential to control ice formation. Credit: Michael B. Davies

IcePic could help atmospheric scientists improve climate change models in the future

Dmitry Spodarets
Dmitry Spodarets

An artificial algorithm called IcePic has been developed by Cambridge scientists. It is capable of outperforming scientists in predicting how and when different materials form ice crystals.

Deep learning is the ability of artificial intelligence to learn to extract information from raw data. It finds its own patterns in the data, freeing it from the need for human input to process the results faster and more accurately. IcePic, in turn, can infer different properties of ice crystal formation around different materials. It has been trained on thousands of images, so it can look at entirely new systems and make accurate predictions based on them.

A foreign material promotes the growth of ice in a film of water. Credit: Michael B. Davies

The IcePic researchers decided to conduct a quiz to test it, in which they asked scientists to predict when ice crystals formed under different conditions in fifteen images. These results were then compared to IcePic's scores. It was much more accurate in determining the material's ability to form ice than more than 50 researchers from around the world. He was even able to determine their errors.

Despite scientists' 75 years of experience in science, IcePic is able to do what human scientists can't.

Source: https://phys.org/news/2022-07-icepic-algorithm-outperforms-humans-ice.html