摘要：AbstractIn this paper, we newly formulate and solve a simultaneous identification problem of nonlinear dynamics and state distribution. This problem is practically useful in many realistic situations, but it has not attracted much attention from the system identification community. From a mathematical point of view, estimation of state distribution is represented as a regularization in terms of the Jensen-Shannon divergence. An important feature of this formulation is its equivalence to the construction ofgenerative models,whose recent progress is one of the most important achievements in the machine learning community. In view of this, we propose an adversarial learning approach, standard technique for generative model construction, to the aforementioned identification problem, and verify its effectiveness through numerical simulation.
关键词：KeywordsNonlinear system identificationgenerative modeladversarial learningmachine learning
