Recursive Feature Machine (RFM)

A kernel-machine probe that iteratively reweights input features via the average gradient outer product (the feature matrix), learning a data-adaptive Mahalanobis metric; used as a probe more expressive than a plain linear classifier while still yielding an interpretable feature-importance direction.