Ok heres the basic idea for the paper.
Id like to present a more or less novel way to represent programs.
A program is a graph. The nodes are typed containers the edges are
functions, whose domain and codoamain match the attached containers
types. The edges of the graph are usable with the usual arrow
operations. That is you can fold the big graph into a smaller one
by applying the arrow operations until you come up with a single
edge. Further the graph is representing a topology and thus you can
glue the graph to other graphs. This is quite novel since usually
in automaton theory you just can concaternate FSMs and such but you
are not able to glue a graph into another or cut it at precise
points. The gule and the folding will allow you to build quotient
spaces on these graphs, which is quite useful if you want to
program with these graphs and also it is quite a nice way to
analyse programs. Also if you have non deterministic graphs, ie
a program that has two possible paths from a source to a sink
you can compare these for runtime speed and choose the faster one
then cap the slower path. All this will make it necessary to
implement an interpreter for the graph programs. The folded 
programs are like compiled programs and would not anymore require
the interpreter. One problem if this approach (the folding) is
cyclic structures of the graph. Implicitly i have the feeling
that category theory is also a nice addition for these graphs.
First i think that they at least form a weak cartesian closed
category and maybe even a topos.

fields this paper would target at would be:
language concepts
concurrent/parallel programming
concurrent/parallel program execution
compilation techniques
automatic program generation
tools and programming techniques