1: \begin{abstract}%  

2: This paper presents Natural Evolution Strategies (NES), a recent family

3: 	of algorithms that constitute a more principled approach to black-box

4: 	optimization than established evolutionary algorithms.

5: 	%

6: 	NES maintains a parameterized distribution on the set of solution candidates, and

7: 	the natural gradient is used to update the distribution's

8: 	parameters in the direction of higher expected fitness. 

9: 	%	

10: 	We introduce a collection of techniques that address issues of

11: 	convergence, robustness, sample complexity, computational complexity and sensitivity to

12: 	hyperparameters.

13: 	%

14: 	This paper explores a number of implementations of the NES family,

15: 	ranging from general-purpose multi-variate normal distributions

16: 	to heavy-tailed and separable distributions tailored towards

17: 	global optimization and search in high dimensional spaces,

18: 	respectively.

19: 	Experimental results show best published performance on various standard benchmarks,

20: 	as well as competitive performance on others.

21: \end{abstract}