Through standard Schur complement calculation, the semi-definite condition can be derived, i.e., The knockoff data are generated by the generator and then sent to the discriminator to verify whether the knockoff condition (Definition 1) holds. The distribution of features w.r .t. samples are shown in Figure S1, and 10K samples are sampled from ImagNet dataset.

This paper proposes a reliable neural network pruning algorithm by setting up a scientific control. Existing pruning methods have developed various hypotheses to approximate the importance of filters to the network and then execute filter pruning accordingly.

We sincerely thank the anonymous reviewers for their support and constructive comments. We will refine the presentation and add a dedicated section for revisiting related-works. More results will be included in the final version. We will refine the presentation around the definitions. These methods re-train the original network to learn the importance of filters.