Consider a set of $k$ continuous variables. Each variable $x_k$ is associated with a hidden distribution from which its value is sampled independently of other variables. I am given a set of observations of the sum of $k$ variables, i.e, $\sum_{i = 1}^{k} x_i$. The challenge is to learn the hidden distribution from the given observations.

This is the general variant of the problem I am trying to study. To simplify the problem, I can make the assumption that the hidden distribution is uniform and of the form $[0, u_k]$. Then, the challenge is to find $u_k$ for each variable $x_k$.

I am not very familiar with learning style algorithms but I am sure variants of this problem have been studied in great detail. Are there any references that I can look at to find out more about this problem?

  • $\begingroup$ Interesting question. In general it definitely seems too hard -- what if one of the variables is itself the sum of independent variables? $\endgroup$ – usul Feb 5 '18 at 1:07
  • 1
    $\begingroup$ Relevant: for the discrete case, you may want to look at the literature on learning Poisson Binomial Distributions, learning $k$-SIIRVs, and learning the parameters of Poisson Binomial Distributions. (Technically, the first two don't learn the hidden distribution, but learn an approximation of the distribution of the sum). See e.g. cs.columbia.edu/~rocco/papers/focs13.html arxiv.org/abs/1505.00662 arxiv.org/abs/1107.2702 $\endgroup$ – Clement C. Feb 9 '18 at 3:46

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.