Protecting Deep Learning: Could the New EU-Trade Secrets Directive Be an Option for the Legal Protection of Artificial Neural Networks?

Deep learning based on artificial neural networks is currently the most promising machine learning method in the field of AI. This paper distinguishes four legal protection objects of artificial neural networks per se: the training data, the topology, the weights as an expression of the trained network, and the specific training method. Both archetypical intellectual property (IP) rights, copyright and patent law, fall to some extent short of protecting these objects. This article examines whether and to what extent trade secret protection could be a suitable or supplementary legal protection tool. Trade secret protection is, among other advantages, flexible. Its greatest weakness, however, is that it allows for reverse engineering which in turn limits its application as a legal protection tool. In the case of an adaptation, trade secret law could at least temporarily supplement patent law and partially replace the classical anthropocentric copyright law in the field of deep learning.