Hacker News new | ask | show | jobs
Show HN: Objective-Reinforced GANs
3 points by gablg1 3297 days ago
A new GAN is in town! A combination of RL + Generative Adversarial Networks that lets you specify objectives for generating data.

For example: "I'd like my generated poems to have fancier words" or "I'd like my generated molecules to have higher solubility." Define simple rewards and ORGAN biases the generative model for you.

arXiv: https://arxiv.org/abs/1705.10843 GitHub: https://github.com/gablg1/ORGAN