[MUSIK]

good afternoon everybody tomorrow I will give again the big picture Monday usually

no big picture just the continuation of the topics we have considered last week

Andreas Maier was my substitute last week and he finished the

topic perceptron we talked about the perceptron the convergence the idea or

the observation that the convergence speed is basically not dependent or

independent of the dimensionality of the feature vectors D then we

talked a little bit about neural networks something that I do not like

so much so the excitement factor is rather rather small and you can

be rather sure that I don't ask usually any questions regarding

neural networks and also the biological motivation as Andreas told you

hopefully last week is not part part of the topic's of

the examinations we usually talk about things that we understand and that

we know and not about biology and all these things

where I only have a vague vague type of understanding

today we have one chapter which might be boring for

you if you are studying mathematics engineering mathematics technical mathematics

because it will be on optimization and we were thinking

a lot about having some kind of refresher course in

optimization or whether we should skip that and assume that

you have all the experience regarding optimization but our personal observation

is that students usually haven't seen so much about optimization in

their career at our university so if it's redundant for you just enjoy

it and hopefully you'll find some weak parts which I do not explain so well so you can

support me and if you see the first time try to catch the basic ideas

all these methods are usually implemented in MatLab and many optimization

libraries so you are not required to implement these methods on

your own but it's important once you want to use it

to have a good understanding what is actually going on in

this or that optimization approach so in general optimization is crucial

for many solutions in pattern recognition image processing usually you can

say what is pattern recognition pattern recognition basically is writing down

an objective function optimize the objective function and then you are

done and this is true for computing the parameters of an

regressor or doing classification we have to optimize with respect to

a decision rule so many of the problems we are considering

are basically reduced to reduced to optimization problems and if you

study optimization in the web in the literature you will see that optimization has

many many different faces you can find whole books whole lecture series

on optimizing and for us as pattern recognition researchers it's very

important also to follow up with current research in optimizations and we

have to say things have changed a lot over the past

10 years over the past 10 years a lot of novel

results with respect to optimization methods were generated and today we

can solve optimization problems where nobody was thinking off let's say 20

years 25 years ago when I was a student so optimization is important very important

if you have a chance to attend lecture on optimization at our university

you should take the chance in particular if you want to do research

with us later on it's always good to have a certain expertise in

this field and it's always good if you have a better expertise in