On Sat, Jun 16, 2012 at 12:18 PM, David Barbour <[email protected]> wrote: > > A valid question might be: how much of this information should be > represented in code? How much should instead be heuristically captured by > generic machine learning techniques, indeterminate STM solvers, or stability > models? I can think of much functionality today for control systems, > configurations, UIs, etc. that would be better (more adaptive, reactive, > flexible) achieved through generic mechanisms. > > Sure, there is a "minimum number of bits" to represent information in the > system, but code is a measure of human effort, not information in general.
I think you'd be interested in this work, http://wekinator.cs.princeton.edu/ The idea is to build electronic musical instruments by training supervised machine learning algorithms on physical input signals (accelerometers, cameras, etc). The machine learning is pretty naive as I understand, but I think it works because the person training the machine is simultaneously exploring the instrument, training herself to play it. Person and instrument learning how to play each other. You can watch an overview video here, http://www.cs.princeton.edu/~fiebrink/drop/wekinator/WekinatorDemo2.m4v Toby _______________________________________________ fonc mailing list [email protected] http://vpri.org/mailman/listinfo/fonc
