This is how Scientists are using Machine Learning to Predict the Unpredictable

You might have heard the butterfly effect, in which a minute change in the complex system can have very large effects everywhere else. According to the butterfly effect, it also suggests that there are some things that even the most advanced science can never predict. But, the list of things that scientists can never predict has now become shorter. Scientists from the University of Maryland have found a way to predict chaos by machine learning.

Reservoir Computing

Generally, when scientists are making predictions about chaotic systems like the stock market or weather conditions, they measure everything as much as possible about it and as accurately as they can, create a computer model, and then see what is done next by the model. But, the chaos theorist Edward Ott and his colleagues were found taking a different approach according to a series of papers published at the turn of 2018 in Physical Review Letters and Chaos. A machine-learning algorithm called reservoir computing was used by them to constantly measure, test, predict, and make changes in those predictions about a chaotic system until they become accurate. The algorithm works with predicting how a wall of fame would behave while moving through a combustible medium such as sheet of paper. It is called the Kuramoto-Sivashinsky equation. It is used to study things like air turbulence and plasma waves. Scientists evolved this algorithm by feeding data from the known evolution of a flame front equation.

Signals are fired by the artificial neurons in the machine-learning network with every bit of data input. Some neurons are measured by the scientists for knowing their signal strength by choosing them randomly. They are then weighted and combined in different ways to have a set of outputs. These set of outputs are then compared with the next inputs. Many changes are made in weights of those signals to improve accuracy of the next measurement. Finally, each and every set is considered to predict how the system will behave. This method successfully predicted the future evolution of that flame wall(nearly eight times) before any other method. According to Natalie Wolchover, you need to measure a typical system’s initial conditions 100,000,000 times more accurately to predict its future evolution eight times.


It is really Unpredictable, But in the end it gets Predicted

There is no doubt that it is very hard to predict such system as no equation is present to describe many of chaotic systems. So, it becomes very hard to make such grand, complex models of such systems. But, machine-learning could easily help to measure behavior in chunks and predict the possibilities very easily. Possibilities such as tsunami predictions, earthquake warnings, and weather forecasts. This method can be used to monitor heart rhythm for impending heart attacks, monitor sun to predict solar storms, and also to predict many such complex systems. Machine-learning can easily fill the gaps of ignorance, as mentioned by Ott. Predicting chaos can be considered as an accurate science and people like James Gleick have played a significant role in this field. James Gleick has written one of the first popular books about chaos theory “Chaos: Making a New Science”.


Please enter your comment!
Please enter your name here