1: \begin{abstract} 

2:   Deep learning requires data. A useful approach to obtain data is to

3:   be creative and mine data from various sources, that were created

4:   for different purposes. Unfortunately, this approach often leads to

5:   noisy labels. In this paper, we propose a meta algorithm for

6:   tackling the noisy labels problem. The key idea is to decouple

7:   ``when to update'' from ``how to update''. We demonstrate the

8:   effectiveness of our algorithm by mining data for gender

9:   classification by combining the Labeled Faces in the Wild (LFW) 

10:   face recognition dataset with a

11:   textual genderizing service, which leads to a noisy dataset. While

12:   our approach is very simple to implement, it leads to

13:   state-of-the-art results. We analyze some convergence properties of

14:   the proposed algorithm.

15: \end{abstract}