1: \begin{abstract}

2: In the classical herding model, asymptotic learning refers to situations

3: where individuals eventually take the correct action regardless of

4: their private information. Classical results identify classes of information

5: structures for which such learning occurs. Recent papers have argued

6: that typically, even when asymptotic learning occurs, it takes a very

7: long time. In this paper related questions are referred. We studiy

8: whether there is a natural family of information structures for which

9: the time it takes until individuals learn is uniformly bounded from

10: above. Indeed, we propose a simple bi-parametric criterion that defines

11: the information structure, and on top of that compute the time by

12: which individuals learn (with high probability) for any pair of parameters.

13: Namely, we identify a family of information structures where individuals

14: learn uniformly fast. 

15: 

16: The underlying technical tool we deploy is a uniform convergence result

17: on a newly introduced class of `weakly active\textquoteright{} supermartingales.

18: This result extends an earlier result of \citet{Fudenberg-Levine-1992}

19: on active supermartingales.

20: 

21: \vspace{1cm}

22: 

23: \noindent JEL classification: D83

24: 

25: \smallskip{}

26: 

27: \noindent Keywords: Social Learning; Information Cascades; Asymptotic

28: Learning

29: \end{abstract}

30: