Okay, first thing, what is a Bayesian network?

Bayesian networks are something many people talk about.

So some people say a Bayesian network is a methodology for representing a full joint probability distribution.

In some cases that representation is compact, meaning small object where we can actually do inference on.

Remember, full joint probability distributions are prohibitively large.

A Bayesian network is a graph whose nodes are random variables, XI, and whose edges denote a direct influence from XJ to XA.

Each node is associated with a conditional probability table CPT specifying blah-de-blah-de-blah-de-blah.

Okay? That's a completely different characterization of Bayesian networks.

It just tells you what the nuts and bolts are.

If you want to define a motorcycle, you can either say that's a device for moving from A to B while experiencing fresh air.

And putting yourself into grave danger, that would be kind of like the first definition.

And the second one would be something like a motorcycle is something that has two wheels and a motor and a seat and a handlebar and a lamp.

And the third one, a Bayesian network is a graphical way to depict conditional independence relations within the set of random variable.

Okay? So this is essentially what does a Bayesian network look like and possibly even what you can do with it.

And of course, Bayesian networks are all of these, and all of these are going to be important.

We want them to be visual so we humans can interact with them.

And we want them to be compact so we have a chance of at inference.

And in order to be that, that's really what they must be made of.

So what you do is you want to look, you want to determine conditional probability of a query variable X given some observed evidence E.

And in that, the Bayesian network encodes our assumptions on the world.

Okay. Russell Norwick has a very nice example, which again shows they're from California.

So somebody buys an alarm. It's a cheap alarm.

And that alarm is not directly connected to the police station.

And so when there's an incident, the alarm goes off and I don't hear it at work.

So I can't call the police. So very easy.

I ask my neighbors to phone me up when the alarm goes off.

Okay. Marion John, my neighbors.

There's a couple of problems with this setup. One is that the alarm isn't reliable.

Sometimes it goes off when there's an earthquake. And when you're in California, there's earthquakes all the time.

You don't even really notice anymore.

Also, there's another problem. John is unreliable and Mary too.

John is a little bit older, so he might confuse the alarm with his telephone and not notice that it goes off or tell me that it went off when he just.

Got a call from his kids and Mary.

Always has loud music on so she might not even hear the alarm.

Okay. Every day situation.

I get a call at work. What is the chance that there's been a burglary?

Okay, it's very important for me to know because I'm going to call call the police and if there's no burglary at the third time, they're not even go to come and check when I call them.

Okay, so I want to be as accurate in predicting what what the chances.

Okay.

Let's try it with logic. Can you dictate, please?

Not really. Okay, too difficult for logic.

All these maybe and maybe not and so on.

Too much too much music for logic here, right? So instead, what are we going to do?

We're going to make a Bayesian network.

And that's something you can.

You can actually.

Making a Bayesian network you can put into a cooking recipe right? First of all, we have to design a set of random variables.

Then we identify their dependencies and then of course we need the conditional the conditional probabilities.

Because I have to know what the probability is of John confusing.

John confusing the.