OptiGAN: Generative Adversarial Networks for Goal Optimized Sequence Generation
One of the challenging problems in sequence generation tasks is the optimized generation of sequences with specific desired goals. Existing sequential generative models mainly generate sequences to closely mimic the training data, without direct optimization according to desired goals or properties specific to the task. In this paper, we propose OptiGAN, a generative GAN-based model that incorporates both Generative Adversarial Networks and Reinforcement Learning (RL) to optimize desired goal scores using policy gradients. We apply our model to text and sequence generation, where our model is able to achieve higher scores out-performing other GAN and RL models, while not sacrificing output sample diversity.
READ FULL TEXT