 I want, this talk is called kittens, moths and thieves, maybe also false advertisements. So let's say I would like to start up, I would, let's say, buy a domain called kittens.com and I have a business plan. My business plan is, well, let's publish good cat pictures. And then I'm going to put ads on the cat content, right, and I'll be rich. This is essentially the business model that Facebook is following as far as I'm concerned. So, okay, what do I have to do if I want to put this into practice? First of all, I need to have good cat content, right? So I need to have good pictures. Let's say this is a good picture. I guess we can all agree this looks cute. I have a question, some little advertisement below it. Maybe people click on it, that's a great idea. If I take this, well, if I make money, I'm going to buy it. So, yeah, somehow I need to generate good content. And, yeah, how am I going to do this? As you all know, the internet is just full of these cat pictures. So how am I going to find the good ones, how am I going to find the bad ones? And we all know machine learning can solve the problem. But now I'm also here at the crypto. I am looking at all these cloud chain docs, and I spend all my money buying A6 and GPUs to mine for coins. So I don't need the computational power to actually do machine learning. What I can do is, maybe in a few years from now, I can go to Google, I can go to Amazon, and we all know these companies do machine learning, right? I guess they are happy to generate the model for me. But they're just going to make happy, right? They sell me the model, but they don't really want me to sell it to someone else. Let's say this company, or one page called Hyperlink.org, I don't know who's behind that. But maybe they're interested in making money as well to fund the research daily, right? So, if I go to Google or Amazon instead, how am I going to do that? This is joint work with people on Facebook. The paper is on watermarking networks. It's a work that allows to watermark these networks. They are for authentication tasks, which is what most people use them for nowadays. It's a generic technique. It comes with a thorough practical analysis of the text, essentially everything you can come up with. Also, yes, which is what I'm talking about this year, is a crypto-like meaning. First of all, it gives you some demographic modeling of the task of watermarking. In a sense, that's described as a game. And also, we describe another task which you could use for machine learning and redevelop production between these two. And to the best of my knowledge and our knowledge is the first time someone describes these four in this context. At the same paper, it's called Turning Your Wings into a Stretch Watermarked in Your Networks in Macdory. It's joint work with your CRD. You can see many paintings from Barry Lange and Mr. Paziz from Facebook.