Hello

and welcome to this unit of our classes.

The topic of today is explainable artificial intelligence.

This is the introductory chapter.

And if you care about our sources

the main source is the top paper.

We have listed here several more papers that are super interesting if you

dare a deep dive into this topic.

So what is explainable artificial intelligence?

Today's agenda is as follows.

First

we are seeing the motivation why this is an interesting topic with the keyword clever hands.

We will see examples such as watermarks on image classification or the cow on beach example

a classic.

Then we will go into the theoretical part.

We will see the definitions of white, gray, and black box models.

And that there's some balance between accuracy and interpretability of the models.

Then we go into the core algorithms.

There are different types of explainer models, explainer algorithms methods.

We will look at the ones that are scoop based like lime

shep

shepley values.

The ones that are complexity based such as division trees or tree pan,

which converts models into decision trees and methodological based.

Here we will take a look at saliency maps

lime one more time

and LRP

layer-wise relevance propagation.

And that is enough for one lecture.

So with that, we will conclude it in the summer.

So let's start with our motivation.

Here you see clever hands

the performing horse with its trainer Wilhelm from Austin.

This is a photograph from 1904 from Berlin.

And clever hands is this beautiful horse

very intelligent as you can see on the left.

You can see numbers and an abacus for doing mathematical operations.

I will point this out with my laser pointer.

Here this is an abacus.

This here is a fancy math and clever hands.

Is this the horse?

So what is this?

Why do I show you a horse?

So the thing is beginning 1891,

the trainer Wilhelm from Austin would demonstrate that Hans is able to do tasks that are almost human intelligence.

And this was done by he asked questions to Hans and Hans would then answer these questions in a variety of hoof taps.

So he would ask Hans to perform arithmetic operations and Hans would answer with the correct tapping of the hoofs.

With an own language.

So Hans the horse demonstrably was super intelligent.

Right?