Vallorie gave a very clear opening description of the problem and motivations.
Basically an object is heated by an external agent. Experimentally you can try to estimate heat transfer to object using an embedded sensor.
In the theory a difficulty is non-unique solutions.
She mentioned the general point that for inverse problem, straightforward physics can still lead to problems that are hard to solve.
A nice quote: "My interest in NKS is to find better strategies for computing solutions to inverse problems".
It was her NKS-2005 Summer School project. The results are in Complex Systems Vol 16, Issue 2. (http://www.complex-systems.com/)
She described the early development of the model. It wasn't obvious how to proceed initially. The original model was a cellular automaton crafted by hand.
Cell values were initially temperature, but a good idea was to model heat values rather than temperature. (i.e. The main physical quantity rather than the observable.)
She learned how to do a search rather than make a model by hand crafting. The resulting search brought up rule 184(?). Then she noticed the model gave some nice features for free.
She figured out how to calculate the thermocouple temperature under the discrete model, which required some thought given the discrete model.
Then she described the search procedure for her inverse problem, which was basically a standard optimization search procedure.
The results for the inverse problem were really good apparently.
She said a big realization was that she could do quantitative modelling with cellular automata, which she had not been convinced of before starting the project.
She also gave a really nice set of practical lessons from the project.
I enjoyed this talk.
Comments