Controlling or regulating a neural network through intermediate latency dreams

摘要

A system controlling or regulating a generative neural network controls or regulates a generative neural network by modifying the intermediate latent space in the generative neural network. The generative neural network contains several layers, each of which generates a set of activation values. An initial layer (and optionally additional layers) receives an input data vector, and a final layer outputs an image generated based on the input data vector. The data that is put into each layer (other than the initial layer) is called data in an intermediate latent space. The data in the intermediate latent space contain activation values (which, for example, are generated by the previous layer or modified using various techniques) and optionally a latent vector. The system that controls or regulates a generative neural network modifies the interposed latent space in order to realize various different effects when generating a new image.