(ORDO NEWS) — The greater number-processing capabilities of artificial intelligence systems mean we can better predict the future of chaotic systems based on fewer and fewer patterns from the past, and a new algorithm makes this process even more accurate.
Developed with next-generation reservoir calculation methods that use a more dynamic and faster machine learning approach, the new algorithm improves the prediction of complex physical processes such as global weather forecasts.
These computational processes, known as spatiotemporal chaotic systems, can now be performed in a fraction of the time, with greater accuracy, using fewer computational resources, and based on less training data.
“This is very exciting because we believe it is a significant advance in terms of data processing efficiency and predictive accuracy in machine learning,” says Ohio State University physicist Wendson de sa Barbosa.
Machine learning is exactly what it is: computer algorithms use a discovery process to make predictions (like future weather conditions) from large archives of data (like past weather conditions).
The reservoir computing approach attempts to more closely mimic the human brain by feeding information into a “reservoir” of randomly connected artificial neurons as a means of discovering useful patterns. The results are then used to inform future learning cycles.
Over time, these systems have become more rational and efficient. One of the innovations in machine learning made it possible to use different components of a predictive model in parallel.
Using this architecture, combined with the latest reservoir computing technology, allows algorithms to identify potential symmetries in what would otherwise be a chaotic mass of information.
The researchers tested their new approach on atmospheric weather models.
Using a regular laptop running Windows software, they were able to make split-second predictions that previously required a supercomputer.
In this particular case, the calculations were made 240,000 times faster than with traditional algorithms.
“If you know the equations that accurately describe how these unique processes for the system will develop, then its behavior could be reproduced and predicted,” says de sa Barbosa.
Machine learning algorithms can be used to predict all sorts of future events, with applications ranging from mundane areas like resource extraction to troubling areas like social engineering.”
As these scenarios become more complex, more and more variables need to be considered, pushing the limits of computing resources.
Machine learning systems are able to detect patterns in past data that the human eye could not detect, and then watch for those patterns to repeat. They may also receive information about themselves to improve their accuracy over time.
In the future, these new and improved algorithms could be used in a variety of situations, such as monitoring patterns, according to the researchers. heartbeat, revealing health issues that would otherwise be overlooked.
“Current machine learning algorithms are particularly well suited to predict dynamic systems by learning their underlying physical rules using historical data,” says desa Barbosa.
“When you have enough data and computing power, you can make predictions with machine learning models for any real complex system.”
Contact us: [email protected]