-
let the output from the pattern recognizer be y1
dark objects and y2 light objects such that
y1 = x1 + x2
-
there is a many-to-one mapping between {x1,
x2} and y1
-
the idea behind the EM is that, even though we do not know
x1 and x2 , we can determine p by
using the knowledge of the distribution
-
Expectation: we compute the expected value of the
x1 using the current estimate of the
parameter pk and the observed data y1
-
similarly the expected value for x2 is given by