New Approach to Machine Learning Could Make Chaos More Predictable

The ability of artificial intelligence systems to crunch a lot of data means we can more accurately predict the future of chaotic systems.

The new method for machine learning improves predictions of complex physical processes such as the global weather forecast.

Simulations of these processes can now be done in a fraction of the time, using less computational resources, and using less training data.

Wendson de sa Barbosa is a physicist from Ohio State University.

Predicting future weather patterns using machine learning is exactly what it is.

The approach attempts to mimic the human brain by feeding information into a'reservoir' of randomly connected artificial neurons. Future cycles of learning are informed by the results.

The systems have become more efficient over the years. There are distinct compoments of the predictive model that occur in parallel. It is possible to spot potential symmetries in a chaotic mess of information by using this kind of architecture.

A new approach was tested by the researchers. They were able to make predictions in a fraction of a second using a normal laptop. The calculations were made more quickly than before.

De sa Barbosa says that if one knows the equations that describe how unique processes for a system will evolve, then its behavior could be predicted.

Machine learning can be used to predict all kinds of future events, finding applications in fields as mundane as mining new resources to those that are alarming.

There are more variables to consider as these scenarios get more complex. Machine learning systems can spot patterns in past data that a human can't, and then watch them repeat. Over time, they can improve their accuracy by feeding themselves.

Monitoring the patterns of a heartbeat, spotting health issues that would otherwise be missed, are some of the things that could be done with the new and improved algorithms.

De sa Barbosa says that machine learning can be used to predict dynamical systems by learning their underlying physical rules.

You can make predictions with machine learning models if you have the right data and computational power.

The research is in a journal.