Putting Sub-solutions Back Together


For the past few years I have been working on a system for evolution
of algorithms for signal understanding (PADO).  I will briefly review
this system and its current state and directions.  I will then spend
the bulk of the talk on the "O" in PADO: ORCHESTRATION.  The general
problem is, if you have decided to try to do divide and conquer on a
problem whose behavior is not well understood, there are a number of
ways to divide, a number of ways to conquer (i.e. recombine /
orchestrate the sub-solutions) and these choices are interdependent.
In the context of PADO, algorithms are trained to solve parts of the
whole problem and then orchestrated into a complete system that
tackles the problem in its entirety.

I will talk about several ORCHESTRATION options and their relative
effectiveness in PADO and generalize this ideas to the general problem
of learning successful multi-agent systems.