Is there a method to calculate Mahalanobis distance incrementally?

112 Views Asked by At

By incrementally (or recursively) I mean to update the pool/group of values/vectors as soon as more vectors are available without having to recompute the entire covariance and inverse matrices.

I thought about something involving Recursive Least Squares to calculate the inverse, but I am not sure this is the right direction.

ps.: I am using it as a dissimilarity measure between two vectors from the pool