Techniques for accurately predicting complex situations known by chaos theory by "machine learning" have been developed

It is impossible to perfectly predict natural phenomena such as weatherChaos theoryIt is known that the condition can change dramatically due to the slight influence that the butterfly fluttedButterfly effectAre also famous. However, research to accurately predict the complex world called "chaos" by "Machine Learning" has been advanced, realizing outstanding prediction accuracy.

Machine Learning's 'Amazing' Ability to Predict Chaos | Quanta Magazine
https://www.quantamagazine.org/machine-learnings-amazing-ability-to-predict-chaos-20180418/

Dr. Edward Otto's research team at the University of Maryland is researching and developing a system that predicts chaos by machine learning. In the study, we use a machine learning algorithm called "reservoir computing" to learn a typical chaotic condition called Kuramoto-Sivashinsky equation.


According to Dr. Otto, the process of machine learning is roughly divided into three. In the example of predicting the chaos of fire spreading method, we first measure the heights of five different places along the burning flame, and this data stream is sent to randomly chosen artificial neurons . In the second step, the neural network learns the movement of the boundary of flames that flares from the input data, weighing the signals in five different ways, and outputting the combined five values, these five Continue adjusting the weights until the numbers match. And in the third step, we actually make a prediction by reservoir who learned the movement of flame.


The feature of Dr. Otto's system is that "reservoir computing algorithm does not understand Kanemoto-Sivashinsky equation at all". In most of the world dominated by chaos, there are many cases where equations expressing the dynamics can not be found, so there is no need to prepare sophisticated equations expressing chaos in the first place, so if you prepare only the solution data, The machine learning system developed by Dr. Otto who can predict is very effective in general chaos prediction.

Mr. Holger Kants of the Max Planck Institute for the study of chaos theory says, "Dr. Otto's paper may be able to predict the weather using machine learning algorithms without using sophisticated models of weather It is suggesting that there is no such thing. "

It is expected that technology that accurately predicts chaos by machine learning can not only detect weather forecast but also signs of heart attack from arrhythmia pattern and predict waves with risk of turning over vessels And that. In addition, Dr. Otto himselfSolar FlareIt seems to want to predict the occurrence.