https://arxiv.org/abs/1907.02544 show that GANs can be harnessed for unsupervised representation learning, with state-of-the-art results on ImageNet. Reconstructions, as shown below, tend to emphasise high-level semantics over pixel-level details.
These results showcase the potential of GANs and other generative models in unsupervised learning, as explained in our recent blog post:
https://deepmind.com/blog/unsupervised-learning/

✴️ @AI_Python_EN