Okay, good.

So in this last lecture of the series, you basically have learned everything there is

to know about neural networks to get started applying them yourself.

And what I'm going to do now in the next 40 minutes or so is just to zoom out and give

you a bit of a grand tour through how machine learning and artificial intelligence can be

applied in the natural sciences, so even beyond physics and what is going on right now today.

And in each of the cases I will, so I have many, many examples that I collected and in

each of the cases I will make a little reference to the kind of technique that is employed.

Sometimes it's a technique that we really learned during this lecture series and sometimes

it's a modification that you are not yet aware of, but that's equally interesting to know about.

So if you think about artificial intelligence and how to use it in the sciences, this is

of course a subject that is much older than the recent developments in neural networks.

And here's a very early example of what people came up with.

This is nothing to do with a neural network, but it's certainly something that you could

classify as artificial intelligence.

So it's using a computer in a kind of creative way to solve complex puzzles that may arise

in the natural sciences.

And in this case, it would be puzzles that arise in chemistry.

So it was a time when NASA was sending space probes to other planets.

The idea was that the space probe might want to land on the planet and then extract rocks

or chemicals from the rocks and analyze those chemicals and maybe look for traces of life

or something like this.

And one of the ways to analyze unknown chemicals is to use mass spectrometers.

And I don't know whether you know how a mass spectrometer works, but basically you take

your complicated molecule, you cut it into pieces, maybe by an electrical discharge or

something so it flies apart into many pieces.

This process is somewhat random.

And then these pieces may be ionized and you can send them through a mass spectrometer

and you find out the masses of those pieces, of those fragments of the molecule.

And you can repeat this many times.

And for a given molecule, it may even fragment in many different ways.

And in the end, you will arrive at a mass spectrum as it is shown here.

So there would be the mass on the horizontal axis and then each peak represents a fragment

that had this particular mass and the height of the peak depends on how many times you

see this particular fragment.

So it's quite a complicated spectrum, so to speak.

And it's really a big puzzle how to put this all back together and figure out what was

the underlying molecule that has fragmented into these pieces.

So it's a tricky combinatorial problem.

And so at the time, these three people and more scientists got together in order to analyze

this problem.

So Feigenbaum came more from computer science and Lederberg and Gerassi were very well known

scientists working in chemistry, for example.

And so they got together and they built a program that would take these mass spectra

and be able to automatically derive how the molecule looks like.

And it would work with a set of rules that have encoded in them the knowledge of experts,

so to speak.

If you see this fragment and that condition is met, then probably this means that the

fragment has a benzene ring, blah, blah, blah, and so on.

So it was at the same time a combinatorial problem, an optimization problem, and also