Wed 5 October, 12:00, WeH 1327


        Response Surface Methods: An introduction, and preliminary
                   discussion of the Auton project.

                          Andrew W. Moore
                            SCS and RI

Response Surface Methods (RSMs) are a heavily used, statistical
technique used in the optimization of (usually) expensive
industrial processes. Such processes are typically characterized
by:

   * A set of controllable parameters (which the Machine Learning
     community would call input variables).

   * A response function, which is a noisy measurement of the result
     of running a process with the given inputs.

   * We desire to find the set of parameters which optimizes the
     expected value of the response.

   * Experiments are very expensive, so in comparison computation is
     very cheap.

This talk will introduce the basic techniques and statistics behind
response surface methods. Current RSM practice is, for sensible
reasons, a far from automatic process.  

Is it possible that techniques from Machine Learning and AI may be
of use in RSM applications? They might permit us to design a new class
of considerably more autonomous systems with which to robustly
enhance, and in some cases even automate, current RSM practice. This
talk will conclude with an outline of a new project to combine a
number of machine learning tools, including memory-based learning and
reinforcement learning, into such a system.