So welcome to the Monday session.

We are currently in the discussion of the expectation maximization algorithm and that's

exactly where I want to hook on.

I avoid now to draw again the big picture because that's not giving so much additional

information compared to last Monday.

So we talk about the expectation maximization algorithm and the question is what does the

expectation maximization algorithm actually do and how does it work?

And for us it's important that we need to identify at least three core components of

our estimation problems.

And these are which ones?

What's your name?

You were too late so you are my candidate.

Stefan.

Welcome.

What are the three components that we have to identify for doing an estimation problem?

The known, known, X, yeah, whatever.

Yes, the observables.

Then the not observables and the entire variables.

And the parameters.

And how do we estimate the parameters B?

Using the EM algorithm.

Yes, how do we do that in the EM algorithm?

By some maximum likelihood estimation.

Yes, what do we maximize?

The Q function according to this here, right?

Yeah.

And our parameter is P naught prime.

And what is the Q function?

Stefan.

It is the, okay, maximum of the integral of P of X given Y, no Y given X.

DX.

Log.

Ah, log, okay, yeah, yeah, yeah.

I'm not here yet.

Yeah, take your time.

Log Y and X with parameters B. DX, Y, yeah, Y of course.

Of course.

Okay, and then we can rewrite this in terms of this probability, how this is equal to?

So P of Y and X with parameters B divided by P of X with parameters B.

Once again?

P of Y and X with parameters B divided by P of X with parameters B.

Right?

No, DY.

No.

DX, okay.

No, Y, sorry, you're right, sorry.

I was so convincing that you trusted me, right?

Why?

Why?

Because we need to marginalize over there.