For tackling the crucial problem described above, we propose here an new version of CL-RPCL, by adding a new
mechanism into FSCL. The basic idea is that for each input
not only the weight vector d, of the unit which wins the
competition is modified to adapt to the input, but also the
weight vector G,. of its rival (i.e., the second winner) is
delearned by a learning rate smaller than that used by Gc.
Specifically,we modify the algorithm given in the beginning
of Section I1 into the following one.
For tackling the crucial problem described above, we propose here an new version of CL-RPCL, by adding a newmechanism into FSCL. The basic idea is that for each inputnot only the weight vector d, of the unit which wins thecompetition is modified to adapt to the input, but also theweight vector G,. of its rival (i.e., the second winner) isdelearned by a learning rate smaller than that used by Gc.Specifically,we modify the algorithm given in the beginningof Section I1 into the following one.
การแปล กรุณารอสักครู่..
