In spite of our 75-year advantage in science, IcePic was still able to accomplish something we couldn’t.
Scientists from Cambridge have created a machine learning algorithm that can outperform humans in predicting when and how various materials would crystallize ice.
IcePic could enable atmospheric scientists in the future in improving climate change models. The findings were published today in the journal PNAS.
Water has certain peculiar qualities, such as expanding when it freezes. Understanding how water freezes around different molecules is important in many areas, from weather systems that influence whole continents to store tissue samples in a hospital.
Water doesn’t always freeze when it’s at zero degrees Fahrenheit, despite the Celsius temperature scale’s premise that it’s the transition point between liquid and solid. Even at -40°C, water can still be liquid, and ice can develop at greater temperatures because of water impurities.
Ice nucleation ability is one of the most important goals of this discipline, and it’s one that has been pursued for many years.
University of Cambridge researchers have created a “deep learning” tool that can anticipate the potential of various materials to generate ice crystals. This technology outperformed scientists in an online “quiz” in which they were asked to estimate when ice crystals would form.
Artificial intelligence (AI) learns to derive insights from unstructured data through deep learning. By identifying patterns on its own, it can handle data more quickly and accurately without requiring human input.
With IcePic, it is possible to infer various ice crystal formation parameters around various materials. IcePic has been trained on tens of thousands of images, allowing it to analyze brand-new systems and make precise predictions from them.
The team created a quiz in which experts were asked to predict when ice crystals will develop under 15 photos depicting various environmental conditions.
Then, these results were compared to how well IcePic worked. When put to the test, IcePic outperformed more than 50 researchers from across the world in terms of predicting a material’s capacity for ice nucleation.
It also helped pinpoint the areas where people were making mistakes.
“It was fascinating to learn that the images of water we showed IcePic contain enough information to actually predict ice nucleation,” says Michael Davies, the study’s first author.
“Despite us – that is, human scientists – having a 75 year head start in terms of the science, IcePic was still able to do something we couldn’t.”
The study of ice production has become increasingly important in climate change studies.
Water is constantly moving throughout the Earth’s atmosphere, condensing to create clouds, and falling as rain and snow.
Different foreign particles, such as smoke from pollution versus smoke from a volcano, have an impact on how ice accumulates in these clouds.
For more precise weather forecasts, it is crucial to comprehend how various factors impact our cloud systems.
“The nucleation of ice is really important for the atmospheric science community and climate modelling,” adds Davies. “At the moment there is no viable way to predict ice nucleation other than direct experiments or expensive simulations. IcePic should open up a lot more applications for discovery.”
Image Credit: Getty
You were reading: IcePic: a New Algorithm That Learns as Fast as Humans – Maybe Faster