Ensembles, Mixtures, and Dyn-Co
In introductory texts that mention combinations of predictors, it is common to draw a bold
distinction between “modular” and “ensemble” systems [51]. This can be regarded as very
misleading, many authors have also commented on the similarities between the two [38, 123],
and as we have seen Dyn-Co seems to provide a link of sorts. In the Conclusions chapter, we
will demonstrate that the technique we study as the focus of this thesis, Negative Correlation
learning [85] provides the missing link that unites these. Though it is not entirely clear how
the dynamics interact, we will show evidence that a smooth space of algorithms exists
between these three paradigms.
Ensembles, Mixtures, and Dyn-CoIn introductory texts that mention combinations of predictors, it is common to draw a bolddistinction between “modular” and “ensemble” systems [51]. This can be regarded as verymisleading, many authors have also commented on the similarities between the two [38, 123],and as we have seen Dyn-Co seems to provide a link of sorts. In the Conclusions chapter, wewill demonstrate that the technique we study as the focus of this thesis, Negative Correlationlearning [85] provides the missing link that unites these. Though it is not entirely clear howthe dynamics interact, we will show evidence that a smooth space of algorithms existsbetween these three paradigms.
การแปล กรุณารอสักครู่..
