1: \begin{abstract}  

2: % AUTOCOUNT-START wc: 167

3:   Turbulence, the ubiquitous and chaotic state of fluid motions, is

4:   characterized by strong and statistically non-trivial fluctuations

5:   of the velocity field, over a wide range of length- and time-scales,

6:   and it can be quantitatively described only in terms of statistical

7:   averages. Strong non-stationarities hinder the possibility to

8:   achieve statistical convergence, making it impossible to define the

9:   turbulence intensity and, in particular, its basic dimensionless

10:   estimator, the Reynolds number.

11:    

12: 

13:   Here we show that by employing Deep Neural Networks (DNN) we can

14:   accurately estimate the Reynolds number within $15\%$ accuracy, from

15:   a statistical sample as small as two large-scale eddy-turnover

16:   times. In contrast, physics-based statistical estimators are limited

17:   by the rate of convergence of the central limit theorem, and

18:   provide, for the same statistical sample, an error at least $100$

19:   times larger. Our findings open up new perspectives in the

20:   possibility to quantitatively define and, therefore, study highly

21:   non-stationary turbulent flows as ordinarily found in nature as well

22:   as in industrial processes.

23: % AUTOCOUNT-END

24: \end{abstract}

25: