Existing methods employ the influence function to achieve feature unlearning, which is impractical for FL as it necessitates the participation of other clients,if not all, in the unlearning process. Furthermore, current research lacks an evaluation of the effectiveness of feature unlearning. Toaddress these limitations, we define feature sensitivity in evaluating feature unlearning according to Lipschitz continuity. Thismetric characterizes themodel output'srateofchange or sensitivity to perturbations in the input feature. We then propose an effective federated feature unlearning framework called Ferrari, which minimizes feature sensitivity. Extensive experimental results and theoretical analysis demonstrate the effectiveness of Ferrari across various feature unlearning scenarios, including sensitive, backdoor, and biased features.

Bayesian predictive intervals are conditioned on the specific observed sequenceZ1:n and make statements on the next value[Yn+1 | Xn+1]. Subjective Bayesian statements on predictions are non-refutable, and are in this sense unscientific, but are optimal according to decision theoretic foundations. However,tomakesuch strong statements, the Bayesian must usually make the strict assumption of the model being well-specified. Asmentionedearlier,computingtheAOI interval is an efficient matrix-vector multiplication, whereas the LOO interval requires expensive broadcastingtoconstructthe ngrid T nISweightarray. We use the same Bayesian model as in (10), again consideringc=1,0.02.

We develop scalable methods for producing conformal Bayesian predictive intervals with finite sample calibration guarantees.

GreedyOur Y with O(n 2) error obtain parameterization (Frank,1956).

We propose a method for knowledge transfer based on a regularization term in our loss function, supervising the sequence of required reasoning operations.