I know it's a seminar with high performance

focus on high performance.

I want to talk a bit about some topics that I think directly go into this direction

but

are not yet on an overly large scale, admittedly.

But nonetheless, I think it's a really interesting and exciting subfield of machine learning

and AI at the moment.

You could summarize it as this generative AI

and I should mention I did get help from

Sora and Co for these illustrations.

You see the dramatic background, but it fits the topic.

So we're trying to actually use similar techniques for simulations

and also a couple of the

images here come from that.

And the main question is, how do we actually use these techniques for physics simulations,

solving PDs?

All right.

I do want to start a bit more general.

And right, this is Sora's interpretation of the challenges of humanity.

So we have quite a few

I think

and we need all the tools we have to address these challenges.

So I used to find it very necessary to motivate why looking at deep learning

AI

machine

learning, neural networks and so on.

Now with the Nobel Prize, the story awards and so on, I think it's clear that there is

something interesting.

It's definitely a powerful technology.

Nonetheless, many open questions.

But I think it's right.

I need to do much less work on defending this combination at all.

So in the context of this generative learning

one of the key drawbacks in a way of traditional

simulations are that they're deterministic.

You could argue that's an adventure.

That's how it should be.

We have a computer.

We need to rely on an input providing a certain output.

If we look beyond scheduling of multiple processes, for example.

But generally, we have a deterministic system, deterministic simulator that gives one input,

produces one output.

The challenge here is that the nature around us is very complicated.

We have ambiguities.

We have uncertainties.

We have effects like the butterfly effect illustrated here, where we're dealing with

certain systems that we do understand.

We can predict how they evolve.

But a tiny change in initial conditions can lead to very large changes later on.

So it's predictable.