Appeal No. 94-4200
Application 07/979,139
sequence utilizing an incrementally adjustable gain parameter.
The independent claim 1 is reproduced as follows:
1. A computer system for machine learning of a time
dependent pattern sequence y(t) comprising:
input means for receiving a plurality, indexed by i, of time
dependent inputs x (t) and a meta-step-size parameter 2;
i
calculation means for calculating from said time dependent
inputs a predicted value, y , of said pattern sequence;*
a computer memory associated with the said means for
calculating;
said calculating means further including a learning rate, k ,
i
exponentially related to an incremental gain $ (t) and a
i
derivation means for deriving the incremental gain $ (t) from
i
previous values of B (t) and having means for
1
Initializing h , a per input memory parameter, to O, and
i
weight coefficients, w , and $ , the incremental gaini i
parameter, to chosen values, i=1,...,n,
Repeating for each new inputs (x ,...,x , y )the steps of:1 n *
calculating,
n
y = E w x
i i
i = 1
calculating,
* = y - y*
Repeating for i = 1,...,n where k is an input learning
i
rate and 2 is a positive constant denoted the
meta-learning rate:
2
Page: Previous 1 2 3 4 5 6 7 8 9 10 11 Next
Last modified: November 3, 2007