Welcome to the second course of this specialization on GANs. In Course 1, you had learned to build a basic GAN, as well as learned many improvements on it, such as DCGAN, WGAN, and conditional GANs. In this second course, you'll learn to take these ideas much further. You'll learn about SalGANs, which are a cutting edge algorithm for synthesizing pictures of faces and other things. You'll also learn about alternatives to GANs, such as the VHV model and other generative models. You'll learn about evaluation of models, including the technical issues of diversity and fidelity. Finally, this course will touch on the important topics of societal implications of GANs, ranging from bias. For example, there was a really unfortunate example of again, taking a lower res image of Barack Obama and turning it into a picture of a white person. The issue of deep fakes. How can we responsibly build and use GANs? With me again is Sharon Zhou, your instructor for this course. Thanks Andrew, and to dive one layer deeper. SalGAN is probably one of those algorithms you've read about in the news. It generates for images that are so realistic. Even people have a hard time telling if a picture was real or a synthetic. There's actually a website called thispersondoesnotexist.com that showcases some of these examples. I remember going to that website and browsing through those pictures. I was actually feeling a little bit sad that all these people they'll never get to meet because they don't actually exists. I've heard actually that people who have started generating stories for these fake people as well. Perhaps, you got to meet them in those stories. Well, SalGAN, will tie into some of the things that you saw in Course 1 about controllability. You get to control how SalGAN will produce various outputs. That can be even as fine grained as taking a wisp of hair and putting it aside like this, which is pretty amazing. SalGAN of course, will be able to generate extremely realistic and diverse images while not collapsing as quickly. Diversity, of course, is important because you don't want to be able to generate just one super realistic face, that's not cool at all. You want to be able to generate a whole range and perhaps even the entire range of possible human faces. Sharon, I know that as one of the leading GAN researchers, you've also put a lot of thought into the social implications, positive and negative of GANs. Yes, I agree. That's why I think that this course interleaves with the societal implications of how we use GANs and also the technical components. As the next generation of GAN builders, I hope with this power of being able to build this amazing style again, you also handle it responsibly, because we want to avoid deep fakes causing havoc. In this course, we will also discuss the very real ethical implications of GANs and when to use them and when not to use them. This course will teach you a relatively powerful set of technologies. Everyone please use these technologies only in a responsible and positive way for society. I think the positive implications actually dramatically outweigh some of the negative use cases of GANs. I really hope that's where you'll concentrate and be using your forces for good. In fact, one of the really cool projects I saw was when you Sharon, were working with Yoshua Bengio and others on using GANs to synthesize images to help people understand the impact of climate change. I hope that many of our learners after taking this course, after taking this specialization, will come up with additional cool creative ideas to help on important problems. I'm really excited to see what you'll create. Let's get started.