I found out that there is really no Step 1, or it is embedded in their algorithm somehow. Supposedly all I have to do is to minimize this equation:

E(x) = sum[ (x-z)^2 ]

Alternating between minimizing with respect to x and to z. This is the step which I don't really understand. To minimize the above equation, I set the derivative to zero and solve, which gave me

x=z

This make sense since when x=z the distance will be minimum, 0. However, it will make no sense since alternating from w.r.t. x to z would not make a difference. And if this is correct, I assume that x=z means that I am just setting the target equal to source at each step, which is what I am currently doing...

Will continue to try...

## No comments:

## Post a Comment