I think what people have learned about genetic algorithms is that the fitness function needs to have some randomness in it otherwise you get something like overfitting.
For example, it might be that you intend to let a robot learn to walk but after running a simulation for 300 hours you find you have taught the robot to walk from point A to point B and the solution doe not work for any two other points. This is a common problem. In any case, a genetic algorithm would be a very poor way to program a machine controller. You want a robust algorithmic solution. But today microcontrollers are so fast, powerful and cheap. The FPGA is not really needed. I can create steps faster than 1MHz with zero measurable jitter with a $2 microcontroller. On Mon, Nov 12, 2018 at 4:12 AM andy pugh <bodge...@gmail.com> wrote: > I remember hearing of this experiment years ago, but hadn't previously > seen the original source. > > > http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.50.9691&rep=rep1&type=pdf > > An evolutionary algorithm was used to create an FPGA configuration to > perform a particular task. > Curiously the final working result depended on a block with no logical > connections to the rest of the network. The solution is dependent on a > quirk of the physics of the actual device. > > -- > atp > "A motorcycle is a bicycle with a pandemonium attachment and is > designed for the especial use of mechanical geniuses, daredevils and > lunatics." > — George Fitch, Atlanta Constitution Newspaper, 1916 > > > _______________________________________________ > Emc-users mailing list > Emc-users@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/emc-users > -- Chris Albertson Redondo Beach, California _______________________________________________ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
