Automated machine learning can classify bound entangled states with tomograms

For quantum systems with total dimension greater than six, the positive partial transposition (PPT) criterion is necessary but not sufficient to decide the non-separability of quantum states. Here, we present an automated machine learning approach to classify random states of two qutrits as separable or entangled even when the PPT criterion fails. We successfully applied our framework using enough data to perform a complete quantum state tomography and without any direct measurement of its entanglement. In addition, we could also estimate the generalized robustness of entanglement with regression techniques and use it to validate our classifiers.