Scientists Use Machine Learning To Peer Into the Future - Latest Tech News

A new algorithm has made it easier to forecast chaotic physical processes.

While the past is an immutable point, machine learning may occasionally help anticipate the future.

Using a unique form of machine learning technique called next-generation reservoir computing, research teams at The Ohio State University have awhile back found a new method for predicting the behaviour of spatiotemporal chaotic systems, such as variations in Earth's climate, that is especially challenging for scientists to anticipate.

The study, published in the publication Chaos: An International studies Journal of Nonlinear Science, employs a novel, super-efficient methodology that, when merged with next-generation reservoir computing, can learn spatiotemporal chaotic systems in a small amount of time needed by traditional machine learning algorithms.

The researchers gave their strategy to the test by forecasting the behaviour of a meteorological weather model, a difficult topic that has previously been intensively explored. The Ohio State team's method is more accurate and requires 400 to 1,250 times less training data than typical algorithms for machine learning that are capable of performing similar tasks. They made predictions in a short amount of time using a laptop running Windows 10, which is around 240,000 times quicker than typical machine learning methods. Their solution is also less computationally costly, whereas earlier, tackling difficult computing issues needed the use of a supercomputer.

"This is quite exciting because we feel it represents a significant improvement in the area of machine learning in terms of both data attentional processing and accuracy rate," said Windsor De Sa Barbosa, lead author and postdoctoral researcher in physics at Ohio State. Learning to forecast these exceedingly chaotic systems, he claims, is a "physics big problem," and that comprehending them may lead to new discoveries and inventions and advancements.

"Modern machine learning methods are particularly well-suited for forecasting dynamical systems by understanding their fundamental physical laws from past data," De Sa Barbosa added. "With enough data and processing capacity, you can predict things about any real-world complicated system using machine learning models." Any physical phenomenon, from the oscillation of a clock's pendulum to power grid interruptions, may be included in such systems.

According to De Sa Barbosa, even cardiac cells exhibit chaotic spatial patterns when they vibrate at an excessively higher frequency than a typical pulse. This discovery might one day be utilised to give improved insight into managing and diagnosing cardiac disease, as well as a variety of other "real-world" issues.

"If one understands the equations that precisely explain how these distinctive processes for a system will emerge, one may recreate and anticipate its behaviour," he added. Simple motions, such as a clock's swinging position, may be easily anticipated given simply its present position and velocity. However, more complex structures, such as Earth's weather, are significantly more difficult to predict due to the large number of factors that actively govern its chaotic behaviour.

To create exact forecasts of the entire system, scientists would need correct knowledge of each of these factors, as well as the theoretical equations that define how all of these variables are related, which is unachievable, according to De Sa Barbosa. However, using their machine learning technique, they were able to decrease the almost 500,000 historical training points utilised in earlier studies for the meteorological weather example used in this study to 400 while still attaining the same or greater accuracy.

"We are living in a world that we currently know extremely little about, so it's critical to understand these elevated systems and understand how to anticipate them more effectively."

Scientists Use Machine Learning To Peer Into the Future - Latest Tech News

Search Tags:

• Latest Tech News
• Trending Tech News
• Machine Learning
• Machine Learning News