Neural networks are considered as one of the most powerful tools for handling a huge amount of data. Thereby, convolutional neural networks are known for their excellent performance in feature recognition.
Those two properties solve the problems appearing in many-body quantum systems and establish methods for solving them in numerically inaccessible regimes. Additionally, researches prove that initializing with transferred features is able to improve the performance of neural networks. In this work, we want to combine both approaches and analyze the effect of transfer learning on neural-network quantum states. Therefore, we create neural networks for classifying representations of 1D and 2D Transverse-field Ising models into their Ising phases. With these pretrained weights as initial values, the performance of predicting ground state vector entries from their corresponding spin configuration is measured and compared to the random initialized case. It appears, that transfer learning is improving the predictions for the two-dimensional case, but not for 1D Ising models.