|
I have a measure of a nuisance signal N in my data X which I want to remove from each sample of my data. I can change N to any value I wish, but I cannot set it to 0 or any fixed value for all my samples because it will produce artifacts in my data. I have P(N|X) from my data and I want to independently re-sample from a new distribution based on the posterior, P(N|X)P(X), which favors smaller values of N. How would I define my P(X)? I know what I want my N's to be, but I don't know how to create a P(X) to reflect this. If I say that P(N|X) is gaussian and P(X) is also gaussian and normalized with unit variance, can I just resample from a shifted version of P(N|X) centered at N=0? |
