Abstract Genetic programming (GP) is a neo-Darwinian search through the
space of expressions to model a given problem. Standard GP suffers from a
problem that its search operators are not well correlated to the behaviour,
or semantics, of solutions. To overcome this limitation, GP researchers turn
to semantic operators, which aim to more accurately drive the search process
through analysis of the behavioural aspects of solutions. A recent
development of this, gaining considerable momentum in GP research, is
geometric semantic GP (GPSGP). In the first part of this talk, I will
discuss the trajectory of research into GSGP, and show how it is essentially
(and unwittingly?) recreating the well-known and understood process of
boosting, but using GP terminology.
The second part of the talk will examine a theory that I have as to why, at least in the field of GP, there appears to be a publication bias against the kind of work that identifies analogies with existing methods, and a corresponding bias towards generating "new" methods. This leads to inefficiencies in research, unnecessary replication of work, and possible problems with the perceived maturity of the field. It will also raise the following question: how did the valid and important scientific endeavour of revisiting the assumptions of previous work become seen as "hostile attacks" on researchers?
I promise that the two topics are related ...
I think ...
Last modified: Wednesday, 05-Mar-2014 10:02:05 NZDT
This page is maintained by the seminar list administrator.