The task of testing whether two uncharacterized devices behave in the same way, known as cross-platform verification, is crucial for benchmarking quantum simulators and near-term quantum computers. Cross-platform verification becomes increasingly challenging as the system's dimensionality increases, and has so far remained intractable for continuous variable quantum systems. In this Letter, we develop a data-driven approach, working with limited noisy data and suitable for continuous variable quantum states. Our approach is based on a convolutional neural network that assesses the similarity of quantum states based on a lower-dimensional state representation built from measurement data. The network can be trained offline with classically simulated data, and is demonstrated here on non-Gaussian quantum states for which cross-platform verification could not be achieved with previous techniques. It can also be applied to cross-platform verification of quantum dynamics and to the problem of experimentally testing whether two quantum states are equivalent up to Gaussian unitary transformations.