From: Ramez Naam (mez@apexnano.com)
Date: Tue Dec 03 2002 - 19:58:35 MST
From: Mike Lorrey [mailto:mlorrey@yahoo.com]
> The problem with conventional software development is that
> all current applications are built by intention according to
> design rules, not by evolutionary processes. A nanoswarm
> design process will give the design for an auto to a neural
> network computer and allow it to evolve the most effective
> cellular differentiation process to achieve the desired
> design within the virtual environment. The most successful
> evolved process becomes the code for the nano-swarm.
Certainly neural networks and evolutionary software design have their
uses and advantages, but they do not address my concern about the
computational cost of simulating these systems. Either an NN approach
or an evolutionary software approach would multiply the required
computational cost by many orders of magnitude.
This is something that's commonly missed by those who propose NNs or
(especially) evolved code to solve a problem. These approaches take
incredible amounts of time to train or evolve, and that time rises as
a steep function of the complexity of the problem you're applying them
too.
cheers,
mez
This archive was generated by hypermail 2.1.5 : Wed Jan 15 2003 - 17:58:33 MST