I have a bonus in a way. I've just found out that Jonas, whom some of you know, is also

a science slammer and has a talk on neural networks of all things in my group. Slightly

embarrassed. I would like to do this before the break and then kind of move on to other

stuff. Okay, Jonas? Oh yeah, there is this thing. You will get your slides. Yeah, that's

about 16 to 9. It's probably all we get. So the idea is that you get a huge audience,

sometimes like 50 people, sometimes like 700 people, and you tell them a bit about... Well,

it's always sold as your own research, but it's mostly something adjacent or doubly adjacent,

something in your field to a general public that is maybe science interested but not necessarily

scientists themselves. So you get a special bonus for this presentation because you already

know all of the actual science. So I skip over the boring parts basically. And the idea

is to get them laughing a bit and to get them science interested and maybe show, hey, here's

a cool thing I can do and just get the population to be a little bit more science into the Essex.

And I think that's important. So please excuse the German text. I knew that I would be doing

this today like 20 minutes before we walked in here, so I didn't have the time to translate

the talk to English. So the talk is about machine learning and why it's maybe more scary

than that sounds. First slide is usually a bit about me. Most of you already have seen

me in some way or another. I'm a PhD student in Professor Kothas's work group, and I'm

a tutor for this lecture. Earlier this year, I still was a master's student at Bielefeld

University. And Bielefeld University has a lot of neural networks stuff going on, so

you can't really study at Bielefeld and be spared this topic in all its egregious detail.

And also coming from Bielefeld, I know that there's some things that we need to talk about

before we can talk about other stuff. So I usually start my science times like this.

If somebody like me is invited to a party, what is usually the first question? Like just

in small talk. This is the part where you shout things. Interaction is allowed. You've

been at these two. Come on. I'm not going to say it's a wrong answer. You can do this.

This is the one part where there's actually literally no wrong answer. Nothing? Okay.

Granted, granted, it was a trick question. People like me don't get invited to parties.

But if I find someone at the coffee machine, that's usually something like, okay, how long

until the robot apocalypse? And my usual answer for the past few years to that has been, oh,

well, as long as Amazon still does the following, oh, Mr. Betzendahl, I see you just bought

a wallet. Can I interest you in more wallets? Then I won't worry about the robots taking

over the world anytime soon. But we need to talk. There's been an interesting paper last

fall that I want to talk a little bit about. And first of all, I don't want to shit talk

machine learning too much. As Professor Kohler mentioned earlier, there's a bunch of interesting

and cool stuff it can do. We're at the point where machine learning can drive cars better

than humans can. And I think when we have like tens of thousands of people dying in

car crashes every day, no, not every day, every year, but still, there's a lot. And

if we can just move everything to the smart cars and they don't drive drunk, that's a

huge gain already. And we also recently got a computer to beat a human in Go, which when

I started studying informatics was told to me would not happen during my time. So that's

happened. I don't want to shit talk it too much. So just so we have this out of the way.

And the next part we can also basically skip because that's what your lecture's been about.

This is a neural network. It works like these neural networks, at least roughly. So rough

approximations so the grant committee will like it. And the general idea, also known

as Hebb's rule, is usually fire together, wire together. The fun is in the, does this

have a pointer? It doesn't. Huh? It does?

Yes. The lower one.

Oh, I see.

And that's recorded, so that's why I would use it.

Okay, makes sense. Okay, so the fun is in all these little connections and how strong