Machine-Learning-Derived Entanglement Witnesses

In this work, we show a correspondence between linear support vector machines (SVMs) and entanglement witnesses, use to generate witnesses for bipartite tripartite qubit (and qudit) target entangled states. An SVM allows the construction of hyperplane that clearly delineates separable states state; is weighted sum observables (``features'') whose coefficients are optimized during training SVM. We demonstrate with method ability obtain require only local measurements even when state nonstabilizer state. Furthermore, SVMs flexible enough allow us rank features, reduce number features systematically while bounding inference error. This derive $W$-state capable detecting fewer measurement terms than fidelity dominant in today's literature. The utility approach demonstrated on quantum hardware furnished through IBM Quantum Experience.