More on reverse modeling


Tom Fiddaman
Junior Member
Posts: 14
Joined: Fri Mar 29, 2002 3:39 am

More on reverse modeling

Post by Tom Fiddaman » Fri Nov 13, 1998 9:51 am

Encoding a production process in an RCA is an extremely interesting idea.
Cellular automata arent really my bag, but Ive only seen examples from
physical or biological sciences where there is an obvious spatial
interpretation. Im interested to hear more about the production process
and how it could be represented as an RCA.

Unfortunately, I think you will discover that running the CA in reverse
will be frustrating, because the results will be extremely sensitive to
initial conditions. The slightest error in encoding the terminal state will
likely lead to unrecognizable results.

The same problem arises with SD-style dynamic models. When you run a model
backwards in time, all the negative loops become positive loops, and it may
be impossible to accurately determine an initial state from a terminal
state. Its easy to model a pendulum, but hard to use a model to determine
where or when a pendulum started swinging by running backwards from the
point at which it has come to rest.

In either type of model, you can use various methods (e.g. hand
calibration, brute force optimization) to improve your estimates of
initial/terminal states or parameters. My guess is that this would likely
be more successful with an SD-type model due to the fewer degrees of
freedom and the greater ability to use uderstanding of the causal structure
to perform intelligent tests.

Again, in the spirit of Jim Hines suggestion that we include reference
modes in our queries, it would help to hear a little more about the nature
of the process and the defective parts.

Best wishes,


Thomas Fiddaman, Ph.D.
Ventana Systems
8105 SE Nelson Road Tel (253) 851-0124
Olalla, WA 98359 Fax (253) 851-0125