yes, but not nondeterministic, just non-uniform in its determinism; the cell replication and death rules depend on the location of the cell>>63373051
fuck chromium, the window magically stopped responding to input so now i have to open a new window and retype my entire response that i don't even remember.
you've got your graph data structure where nodes represent neurons and edges represent synapses. that's a markov model. even though you're following all possible paths at once and not randomly selecting each edge based on its weight, it's still a markov model, because the edge weights still basically represent probabilities. not the probability that the edge will be selected, because unless the strength of the preceding signal is 0, it definitely will -- but the network's "perceived" probability that following the edge will have been the right choice -- that is, the certainty.
then superimposed on that graph you've got another graph data structure, although you may very reasonably choose not to store it that way: you've got the state of the breadth first traversal. you do a breadth first traversal through the markov model, summing up edge weights leading to each node as you go along and then scaling the result by the properties of that node, and what do you have at each stage of that process, you have a set of signals with different weights, and each of them is somewhere different in the graph. so you have things that have state, have position in a data structure, and use this information to decide whether to propagate and/or die. sure sounds like cells to me. so then you've got your cellular automaton.
therefore a neural network is a cellular automaton over a markov model