Alright, welcome to Exercise 3, an introduction to machine learning.

And today we will do some exam prep.

So we will take a look at a past year's exam, write a solstrum together, and hope that you

will learn something from it and that it will be a nice preparation for you for this year's

exam.

This year's exam will be fully MC based, so multiple choice based, so it will differ from

the past year's exams, which also had lots of free text questions.

But don't be discouraged, this year's exam still won't be too hard and as you will see

from the questions that I show you, it's usually quite easy to answer if you are somewhat well

prepared.

Alright, without further ado, let's start.

So here we have for example the question, what is the most important requirement for

supervised learning?

Like the best answer.

So this can be or could be a question and yeah.

If you look at it, you might get a bit discouraged, maybe not in this case, but in some cases.

But if you look at it, a continuous range for the quantization, is that important for

supervised learning?

Is it even relevant for supervised learning?

And no, it's usually not.

No, I'm just saying no, that's not it.

High dimensional feature vectors.

Well sometimes high dimensional feature vectors can even be detrimental.

We don't want feature vectors with too high dimensionality because of the curse of dimensionality.

So let's discard this one.

Label training data that looks good.

A high signal to noise ratio.

Always nice to have, but do we need it?

Is it the most important requirement for supervised learning?

More important than having label data?

I think so, so we can discard it.

And finally, low intra-class variance.

Well that can be good, but it's also not as important as label training data.

So we discard it and then we have found our optimal solution.

Not that difficult I assume.

Alright, with multiple questions like this, here for example, you are asked about the

four different types of learning paradigms, supervised, weekly supervised, semi-supervised

and unsupervised, and then you should match it with the different types of available data.

That's also a type of question that you might be asked in the exam because it's quite easy

to check, to mark.

So prepare for questions like this and it's also very easy.

What is supervised?

What type of data does supervised need?

Obviously fully labeled data.

Weekly supervised needs points, scribbles, so partially labeled data.

Semi-supervised learning is a combination of self and supervised learning.

So self-supervised learning and fully supervised learning and this needs labeled and unlabeled

data or labeled and labeled.

I think that's correct.

And finally unsupervised learning needs unlabeled data.