The present disclosure relates to systems and methods for using transfer learning in log parsing neural networks. In one implementation, a system for training a neural network to parse unstructured data may include a processor and a non-transitory memory storing instructions that, when executed by the processor, cause the system to: receive unstructured data; apply a classifier to the unstructured data to determine that the unstructured data comprises a new category of unstructured data; in response to the determination, identify an existing category of unstructured data similar to the new category; based on the identified existing category, select a corresponding neural network; reset at least one weight and at least one activation function of the corresponding neural network while retaining structure of the corresponding neural network; train the reset neural network to parse the new category of unstructured data; and output the trained neural network.