4
$\begingroup$

To put it simply, what if the sample space underlying our probability space is a category instead of a mere set. Has a theory or probability and statistics been developed for such a situations in which the samples can have relations between each other?

This questions is somewhat similar to this one however I do not find the answer to my question in there.

Improve this question
$\endgroup$
3
  • 1
    $\begingroup$ Would you be satisfied with a groupoid instead of a category? I don't know of any such development offhand, but this case seems more likely if it has been done. $\endgroup$
    – David Roberts ♦
    Commented Jul 12, 2020 at 11:40
  • 2
    $\begingroup$ There is a highly upvoted earlier question that asks for an introduction to probability theory from a categorical perspective: mathoverflow.net/questions/20740/… $\endgroup$
    – Mark Wildon
    Commented Jul 12, 2020 at 13:10
  • $\begingroup$ Posets are the special case of categories where there is most one morphism between any two objects. See e.g. sciencedirect.com/science/article/pii/S0020025519311338 for one recent paper on random posets. $\endgroup$
    – Mark Wildon
    Commented Jul 17, 2020 at 18:40

1 Answer 1

Reset to default
4
$\begingroup$

REVISED ANSWER

  1. First Approach, from Random Graphs

Never seen it anywhere. Anyway, here is my two-cents:

First of all, there is a well established theory of RANDOM GRAPHS (see here for a ref. Of course, there are many choices available, you may make nodes, edges or both random, and you may confine yourself to certain classes of probability distributions).

Now, a a category is a (directed) graph, in the precise sense that there is an obvious forgetful functor $ F: Cat \rightarrow Graph$.

The first step toward defining a random category would be to say that it "forgets" to a random graph. In other words, just like the category of (deterministic) Graphs embeds as a subcategory of Random Graphs, the category of standard deterministic categories CAT would embed in the cat of Random Categories, in such as way as to preserve the respective forgetful functors.

A category has more structure than a directed graph, it has composition, identity, and also commuting diagrams, so you need to add those to the picture:

composition is easily taken care of, for instance by stipulating that the probability of a composite arrow is the product of the probability of its components. Similarly, identity morphisms could be assigned the same probability of their corresponding nodes.

Now you are left with the diagrams (I am talking now of only those diagrams which are not forced by the laws of composition). There you have a certain degree of freedom. My intuition would be: first play with random free categories constructed from random graphs by free completion, then stipulate an assignment of probabilities to the space of all generating diagrams in the category.

  1. Second Approach, from Random Simplicial Sets.

Another (broader) take on the above: categories are, in a sense, particularly simple types of SIMPLICIAL SETS, namely the ones which satisfy the horn condition.

So, rather than starting from random graphs and building your way up to categories, start with a notion of random simplicial set and by restriction you will have your random categories.

How? Don't know if the notion of random simplicial sets has been explored, but certainly there is something on random simplicial complexes, see for instance here). This could be used as motivating example.

Hope it helps

PS Final comment: in fact approaches 1 and 2 are in the end the same, 1 is part of 2, because graphs are simplicial sets of dimension 1.

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ Thank you. Actually random graphs would wotk for me too, I just need a model with variable numbers of nodes and edges at the same time. $\endgroup$
    – Reza Rezazadegan
    Commented Jul 14, 2020 at 4:54

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .