Machine learning predicts extreme events in ultrashort pulse lasers

TL;DR

This paper demonstrates that a nonlinear neural network can accurately predict extreme events in ultrashort pulse lasers, achieving high accuracy with experimental data and perfect prediction with theoretical models.

Contribution

It introduces a neural network model capable of predicting extreme laser pulse events with high accuracy, bridging experimental and theoretical data.

Findings

95.45% hit rate with experimental data

100% prediction accuracy with theoretical data

False positives are 6.67% experimentally and 23.33% theoretically

Abstract

In this paper we present a nonlinear autoregressive neural network with a hidden layer of 50 neurons, three delays and one output layer that accurately is capable of predict the appearence of extreme events in a Kerr lens mode locking Ti:Sapphire laser with ultrashort pulses. Extreme events are produced in the context of a chaotic atractor and with chirped pulses. The prediction of this neural network works well with experimental and theoretical time series of amplitude of laser pulses. When fed with experimental time series we have 95.45\% of hits and 6.67\% of false positives while using theoretical time series the network predicts 100\% of extreme events but the false positive rise to 23.33\%.