-- Version 1 of the paper was entitled "Nearly Optimal Learning and Sparse Covers 
   for Sums of Independent Integer Random Variables"
   and was published on arxiv on May 4, 2015.


-------------------
Version 2 Changes:
-------------------


-- Changed the title, revised abstract and intro.
-- Added new Fourier-based algorithm that has optimal sample complexity (up to a constant), 
   improving on the algorithm from Version 1 by logarithmic factors.
-- Added sample complexity lower bound for any value of k. Version 1 had an optimal sample lower bound for k=2.
   A generalization of the k=2 approach gives an optimal lower bound for all values of k.
-- Added argument that the succinct description of the hypothesis, via its DFT, is efficiently samplable.
-- Added simple generic algorithm that applies for any distribution with a sparse FT.