So welcome everybody to our first lecture of deep learning this semester. My name is

Tobias Wöhring and as you probably noticed, so if you look at who are we, well the lecture

is actually given by Professor Andreas Meyer but unfortunately cannot be here today so

I will be your substitute. This is me. I will be your substitute today and I'm supported

by my colleagues Leonie, Georgina and Nishant which you can find over here, here and there.

And the exercises which you all have to attend if you want to actually do complete the module

and we will be supported by Helin and by Helix who are also sitting over here.

So next lecture Professor Meyer will probably be back so you will see him there.

So let's start off by introducing some words which you will find flying around deep learning.

So deep learning has become a topic of huge interest for a lot of people and you see popping

up many words in conjunction with it. So like there are different tasks which are nowadays

tackled with deep learning and like the most classical ones are classification where you

want to find a label for example an image or any particular observations you want to

label and classify. So my classic example of this you have images

of cats and dogs and you want to have an algorithm which actually tells you if an image shows

a cat or a dog automatically. So that's classification, a discrete label

for an input and opposed to that is regression where you actually want to have a continuous

valid output for a given input. Now those are the two most classical topics.

Now there are actually applications of those two to different more application specific

tasks. So for example there is the task of segmentation

in image processing where you want to extract the outline of an object or the special task

of generation which here refers to having a generative model, so a model of a process

which tells you how stuff is generated and actually allows you to draw samples from it

and therefore for example sample new images, ever unseen images which the neural network

in this case has created. So those are tasks nowadays associated with

deep learning and there are many many more but I think those are the most prominent ones.

Now there is also other words which are very related to deep learning because people use

them all the time like big data and artificial intelligence. Those are probably the most

important word which surround the hype around machine learning which has really occurred.

So big data is actually an interesting one because nowadays everyone says that deep learning

is very related to big data. But in the beginning when actually this topic

of big data was introduced it was mostly not only referring to data which consumes a lot

of memory but also on data which is differently structured than image data.

So what am I talking about? I am talking about search engines. So you have tables of different

linking structures and you have terabytes of that and this was actually the original

field where this big data technologies emerged. But nowadays mostly the property of huge memory

consumption is associated with big data and therefore it has become closely related to

deep learning for some reason. The other name is actually artificial intelligence

and it is actually the very broad super topic to actually machine learning and deep learning.

So artificial intelligence refers to a very I would say not so clearly defined field where

you want to have intelligent behavior for some computer program.

One thing which is certain about artificial intelligence that part of it is being able

to teach a machine how to learn stuff, how to acquire new knowledge and extract it from

past experiences. That is actually the subfield of artificial intelligence which is called

machine learning. Now machine learning or pattern recognition

is actually the super field of the field we want to study this semester which is called

deep learning. So all the deep learning techniques are particular machine learning techniques

and there are also many other machine learning techniques which probably some of them you

should have encountered in prior lectures. The last name here is actually representation

learning which I will come back in a minute. Associated with deep learning there are also