OK, welcome back to AI 2.
Oh, let's see.
That's better, probably.
We were talking about probabilities yesterday,
starting to build up the machinery that
lets us talk about reasoning under uncertainty
and acting under uncertainty, which
is exactly what we want our agents to be
able to do this semester.
And before we go into the maths, which is rather simple,
and you've probably heard it before,
I would like to make sure that we're
on the same page in what we're really doing here.
And the upshot of everything I tried to say yesterday
was that we are using probabilities
to model incomplete knowledge.
So it's not that the world is uncertain or something
like this as an object per se, but it's we who are,
or the agents who are uncertain about the actual state
of the world.
We're actually trying to model our own limited or the agents'
limited knowledge.
The world itself only has one state.
OK, we just have no idea what that is,
or we have a limited idea of what that might be.
And when we're looking at things like likelihoods
or probabilities, we're really interested in,
of all the possible worlds, how many are in that state
that I'm using to model as a possible world?
In all the possible worlds out there, only one of them
is the actual world.
How many are consistent with cavity equals true?
And there's a slight question there,
which I would like you to think about,
is that the actual possibilities in the world
might actually be quite endless.
We're only caring about a couple of parameters there.
We're only distinguishing certain worlds.
When I'm wondering about, and I only
have limited information about, the weather outside,
it's probably sunny because it was five minutes ago,
but who knows?
Then, of course, the world, there
are a lot of possibilities.
And of course, the world, there are lots of possible world
states in ways that are uninteresting to me.
For instance, exactly how many Chinese people there are.
Not interested in that.
In the moment, I have a rough number there,
but exactly how many there are is completely uninteresting
Presenters
Zugänglich über
Offener Zugang
Dauer
01:22:10 Min
Aufnahmedatum
2018-04-19
Hochgeladen am
2018-04-20 10:42:48
Sprache
en-US
Der Kurs baut auf der Vorlesung Künstliche Intelligenz I vom Wintersemester auf und führt diese weiter.
Lernziele und Kompetenzen
Fach- Lern- bzw. Methodenkompetenz
-
Wissen: Die Studierenden lernen grundlegende Repräsentationsformalismen und Algorithmen der Künstlichen Intelligenz kennen.
-
Anwenden: Die Konzepte werden an Beispielen aus der realen Welt angewandt (bungsaufgaben).
-
Analyse: Die Studierenden lernen über die Modellierung in der Maschine menschliche Intelligenzleistungen besser einzuschätzen.