Ohio State University researchers are using a new type of machine learning technique called next-generation reservoir computing to predict the behavior of spatio-temporal chaotic systems that are particularly complex for scientists, such as changes in the Earth’s weather. I recently discovered a method. Predict.

A study published in the journal today Chaos: An Interdisciplinary Journal of Nonlinear Scienceutilizes new highly efficient algorithms that, when combined with next-generation reservoir computing, can train spatio-temporal chaotic systems in a fraction of the time of other machine learning algorithms.

The researchers tested algorithms to predict the behavior of a complex problem that has been studied many times in the past: atmospheric weather models. Compared to traditional machine learning algorithms that can solve the same task, the Ohio team’s algorithm is more accurate, using 400 to 1,250 times less training data to make better predictions than the corresponding team. Their method is also computationally less expensive. Solving complex computing problems used to require a supercomputer, but with a laptop running Windows 10, we made a prediction in about a second. This is about 240,000 times faster than traditional machine learning algorithms.

Wendson De Sa Barbosa, lead author and postdoctoral researcher in physics at Ohio State University, said: Learning to predict these highly chaotic systems, he said, is a “grand challenge for physics,” and understanding them could pave the way for new scientific discoveries and breakthroughs. said that there is

“Modern machine learning algorithms are particularly well-suited for predicting dynamic systems by using historical data to learn the underlying physical rules,” said De Sa Barbosa. “Given enough data and computing power, we can use machine learning models to make predictions about complex real-world systems.” It can include any physical process, up to and including disruption.

Even heart cells exhibit chaotic spatial patterns when they vibrate at frequencies abnormally higher than normal heartbeats, said De Sa Barbosa. This means that this research could one day be used to provide better insights for controlling and interpreting heart disease and other ‘real world’ problems.

“If we knew the equations that describe exactly how these unique processes in a system evolve, we could reproduce and predict their behavior. Simple movements like the swing position of a watch. can be easily predicted using just the current position and velocity, predicting a more complex system like the Earth’s weather, where many variables actively determine its chaotic behavior, is much more difficult.

To make accurate system-wide predictions, scientists need to have accurate information about all of these variables, and they need model equations that describe how many of these variables are related. But De Sa Barbosa says this is simply not possible. However, with their machine learning algorithms, the nearly 500,000 historical training data points used in previous studies for the atmospheric weather example used in this study can be reduced to 400 and still perform the same or less. We were able to achieve higher accuracy.

Moving forward, De Sa Barbosa aims to take his research even further by using algorithms to speed up spatio-temporal simulations, he said.

“We live in a world that is still largely unknown, so it is important to recognize these highly dynamic systems and learn how to predict them more efficiently.”

A co-author of the study was Daniel J. Gauthier, a professor of physics at Ohio State University. Their research was supported by the Air Force Office of Scientific Research.

Story source:

Materials provided Ohio State UniversityOriginally by Tatiana Woodall. Note: Content may be edited for style and length.