https://arxiv.org/abs/2010.15040
Training Generative Adversarial Networks by Solving Ordinary Differential Equations (Chongli Qin, Yan Wu, Jost Tobias Springenberg, Andrew Brock, Jeff Donahue, Timothy P. Lillicrap, Pushmeet Kohli)
gan 학습의 불안정성이 discrete한 update 때문에 발생한다는 추론. continuous update로 학습 과정을 대체하고 역시 ode가 출동.
#gan #neural_ode