Introduction to Machine Learning

Video thumbnail (Frame 0) Video thumbnail (Frame 9260) Video thumbnail (Frame 13134) Video thumbnail (Frame 15737) Video thumbnail (Frame 26967) Video thumbnail (Frame 30086) Video thumbnail (Frame 41945) Video thumbnail (Frame 47543) Video thumbnail (Frame 49764) Video thumbnail (Frame 53900) Video thumbnail (Frame 54815)
Video in TIB AV-Portal: Introduction to Machine Learning

Formal Metadata

Introduction to Machine Learning
Title of Series
Part Number
Number of Parts
No Open Access License:
German copyright law applies. This film may be used for your own use but it may not be distributed via the internet or passed on to external parties.
Release Date
Production Year
Production Place
Dubrovnik, Croatia

Content Metadata

Subject Area
Keywords machine learning
Point (geometry) Presentation of a group Group action Link (knot theory) Algorithm Decision theory Virtual machine Gene cluster Mereology Number Data model Inclusion map Goodness of fit Unsupervised learning Performance appraisal Machine learning Fluid Cross-correlation Read-only memory Semiconductor memory Term (mathematics) Energy level Software framework Extension (kinesiology) Social class Area Machine learning Artificial neural network Bit Database Measurement Category of being Fluid Process (computing) Telecommunication Software framework Right angle Game theory
Area Slide rule Dialect Direction (geometry) Visual system Planning Bit Contingency table Evolute Measurement Number Fluid Mechanism design Read-only memory Natural number Robotics Video game Software testing Process (computing) Quicksort Endliche Modelltheorie Position operator
Group action Transportation theory (mathematics) Multiplication sign Decision theory Source code Artificial intelligence Water vapor Mereology Collective intelligence Computer programming Formal language Mechanism design Mathematics Sign (mathematics) Computer cluster Hypermedia Core dump Kognitionswissenschaft Endliche Modelltheorie Physical system Area Boss Corporation Arm Network operating system Bit Lattice (order) Arithmetic mean Process (computing) Website Right angle Quicksort Physical system Rational number Game controller Observational study Virtual machine Student's t-test Number Goodness of fit Root Natural number Software testing Condition number Form (programming) Computer architecture Artificial neural network Contingency table Particle system Film editing Integrated development environment Function (mathematics) Gradient descent
Group action Graph (mathematics) Outlier Decision theory Multiplication sign Set (mathematics) Water vapor Insertion loss Function (mathematics) Computer programming Dimensional analysis Computer Neuroinformatik Medical imaging Machine learning Different (Kate Ryan album) Object (grammar) Physical system Exception handling Area Machine learning Algorithm Theory of relativity Outlier Computer Sound effect Measurement Virtual machine Category of being Data stream Digital photography Arithmetic mean System programming Self-organization Normal (geometry) Pattern language Right angle Procedural programming Figurate number Task (computing) Resultant Row (database) Point (geometry) Web page Slide rule Inheritance (object-oriented programming) Algorithm Computer-generated imagery Virtual machine Similarity (geometry) Event horizon Rule of inference Number Wave packet Latent heat Goodness of fit Representation (politics) Traffic reporting Task (computing) Rule of inference Pairwise comparison Information Projective plane Computer program Expert system Code Database Cartesian coordinate system System call Exploit (computer security) Event horizon Video game Game theory Musical ensemble Object (grammar) Table (information)
Point (geometry) Pattern recognition Slide rule Group action Multiplication sign Computer-generated imagery Visual system Gene cluster Set (mathematics) Similarity (geometry) Distance Rule of inference Neuroinformatik Local Group Medical imaging Unsupervised learning Machine learning Hypermedia Different (Kate Ryan album) Pattern language Ranking Data structure Endliche Modelltheorie Pairwise comparison Error message Task (computing) Social class Area Scale (map) Curve Software bug Algorithm Distribution (mathematics) Pattern recognition Software developer Projective plane Line (geometry) Cartesian coordinate system Measurement Uniform resource locator Loop (music) Website Right angle Object (grammar) Ranking Species Spacetime
Metre Group action Algorithm Source code Merkmalsextraktion Control flow Set (mathematics) Bit rate Mereology Hypercube Formal language Word Data model Medical imaging Latent heat Maschinelle Übersetzung Configuration space Endliche Modelltheorie Machine learning Parameter (computer programming) Line (geometry) Virtual machine Population density Error message System on a chip Configuration space Object (grammar) Arithmetic progression
Cluster sampling Multiplication sign Set (mathematics) Database Software bug Data model Type theory Labour Party (Malta) Different (Kate Ryan album) Object (grammar) Office suite Endliche Modelltheorie Error message Partition (number theory) Social class Machine learning View (database) Range (statistics) Open set Virtual machine Partition (number theory) Process (computing) Self-organization Pattern language Right angle Reduction of order Spacetime Point (geometry) Web page Virtual machine Control flow Rule of inference Performance appraisal Unsupervised learning Population density Data type Window Distribution (mathematics) Slide rule Forcing (mathematics) Database Similarity (geometry) Inclusion map Wiki Personal digital assistant Sheaf (mathematics) Object (grammar) Videoconferencing Form (programming)
a good. similar to mark all the the horny i also don't know about your bigger ounce because i think me a few are already advancing machine learning but i don't know if you've everyone that's why i would like to ask you who actually here's an idea about machine learning what if these in general.
ok many people would if. still only classes. and i will see how i can surprise you have a good who had deployed in classes. and who applied the machine learning the real task. polls but not everyone right and the it is it is also impossible in three hours or what they were is left to would you got a little action to learn about an area in the property what machine learning and idea of this nation is just that the for those of you who see the machine learning for the first one not most. familiar machine learning just to have an idea in the daily work if they see a problem to think ok i might be able to solve these was motioning i might applied here and they get the rough inclusion how it works because many people who didn't have classes but up a plank machine learning to use the ages. apply them because they work but maybe the intuition the behind is not known and to these another are over fifty five and the with a hands on you can also of where it will be of practical experience with those small two books were. you can to some of the some poured actually and and change some parameters and see what happens and of course the the one important point is to have the common base for discussions because you mean your few up light machine ago or use machine learning more about machine learning but the there is no all come on base. it probably is like an elephant in the dark room everyone that should from different sides during the next week the the we hope that we can have a common based on the framework an exercise this the links are here but actually an eric is supporting here and he will. the. he will possibly was be sticks and also the the biden old books. ok. so i was thinking then the i was expecting actually that mean you if you already have some experience in machine learning so i would like to start very very genome very high level maybe were where everything comes from and why it's so interesting so we will start with the introduction of tuition diligence and the then we. we will see submissions on in calgary games for answer was ordering for super was warning and actually learn a little bit how we evaluate more those that are produced by these methods and finally eric will be for. from your presentation of all the cutting edge research in this area. but first for people with the lord is a part of artificial intelligence and if we talk about some think we should define it a little bit rights live with the of common understanding. so i would be interested in what do you think what is actually intelligence. can you just name some aesthetics or would do think what what what is intelligent what is not intelligent. reasoning year. the community facilities. making decisions so you that probably you see someone to digital age and you will recognise that the these intelligent right but generally these are all aspects is again is an elephant in the room it's hard to define it very kreis what is the end will see also some examples. and there was just him and the whom you probably know a few what little bit with that the six the relies on measurements for correlation another's is and he he's the one who came up with a cutest so just with the us on the instrumental measure did the intelligence and he divided. of all he found five as they actually have are often to virgin's is like reasoning the it was said here already a fluid reasoning also a is that ok twenty five he presented his working with our number of medical problems then fluid reasoning is a is a flexible solving problems or dropping from you. problems with measles special processing. think it's it's clear then we're one should have a long term memory and the short term memory long term memory like a database right in the short term memory like for making fuzzy seasons and now another question to you. what do you think is more intelligent guy so a particular human who is especially since one of these aspects or one who has abilities everywhere equally was more intelligent. everywhere. what the it turns out that it is partially like in between right you can have the you can of the year a very intelligent everywhere. but the you can be very diligent the couple of the suspects similar to what is so core player so far sportsman right he can play soccer but he may not swim very well but he will scream may be better than ever rich guy right so yes some abilities this is what spearmon actually found out that the someone who. this would some of these. clusters is also would in other but to actually look not there for free. these are the small is already quite group the you what have they did actually buy for governor in the eighty three he said if he is the extent of the it also review for example he said the interpersonal communications feels it also part of the intelligence.
of human and the not not just the reasoning and more each day. good for now. is it is a year later little we know a little bit defined the intelligence a little bit and all the question is what are you have to do it. or. if you're it's worth it. we know we we get we can discuss discuss these indeed it's the measurement is not perfect and there's also of criticism towards it. for example if someone can or cannot work with numbers like some some of it all. them. some of the region or so some people who never worked with numbers this you intelligent of course but they will help us that this for example write about mobile them and they were thinking about their switch which would capture even even these and it came with some driving tests all the more details you draw actually the antigens you are. and then they found there are still are some some people in africa will just never drawl so the they don't have these in the culture and they couldn't us even the statist but they could do other things that are right the these the the cutest of course is just just an instrument is not the it will not merely about the intelligence. since the just it's just as the mission or whatever you want to. with and yet so now my position is the do you think that's the artificial intelligence is to exist. it all so what now we have with you find little bit of human intelligence. what about the future intelligence who thinks that were kept artificial intelligence so that it yet. ok we will see if we know we use it everyday actually in its area some some aspects of it. and yeah let's just go to the next slide or the seas i feel sketch how our technology actually the dust evolution or so were first absorption of sort of nature like works or the birds or of moving the anymore.
as and then we got came up with a very simple model which was doing a particular think so it was and what they like itself. exactly stopped doesn't soul for their the wheel doesn't solve the general problem off of moving around but it's very efficient the with one particular movement one particular direction and also the airplane their plane is a is much faster than any flying anymore but with it in or it is a flexible. as it is the is a bird and once we come up with a small more those the to solve the problems we can to get more knowledge about the problem and allies it and come up and with that earth solutions further like the now there was a robotics have also worked on the bitter been so the all of the robots. and we can drawings which are actually very flexible in flying into can was shun them in particular the or location. so it's getting better but with this is a simple others also to understand more about the problem and the now the question we absorb some of us wanted to know just how should the more the real life for fourteen to regions and how it will develop. and actually yeah what is what is the mechanics you can explain the intelligence of absorption the nature for looking in the nature we see here for an example so contingents a hear what the numbers are here for get here different animals and.
the numbers. for. year expected so it's a couple of times not everyone immediately says the neurons but these are morons an example of contingencies of course the human being it doesn't mean that the human is the only possible intelligence maybe they will be some aliens in will shoal as things that we don't know which also. religion but this is one that we can absorb and and study them and it is around the eighty six uga neurons right and then we can hear the ape which is ok we'll also see the elephant you and it has more but the it is not intelligence. is not as it continues as human that means that the number of neurons is not everything is just an eye. or. oh. but it. you know. for millions. it can play. and yet they have a remit is smart the most also they can roll right so that you can teach them to draw pictures like like a human to the control themselves of course they don't grow like they don't know what they're doing but you can train them they remember all the movements and can actually replicate the drawing. but. i know where or and i were an hour. they're also a good. they will talk about mark is it the was a somewhat about them as well and so why it could hear the chimpanzee because it costs them your test very well so the new york this is like the test if they know can recognize itself in the mirror than a boss is so there are different approach. crystal tested and with the ape you even don't need to put marker something you know that you recognize itself right then the birds birds use students were also know if they'd also some of the worst possible so the the new york test model monkeys actually possible to us but you been easy to do. the birds have just some of them pass and they used with the with five hundred millions oath of neurons in the brain and the awful posting your aunt's has the red two hundred fifty media and you know if it can it get some community for the meeting has been confined of the way from leverage and so on. and then we descent we see here there's ever a fish which is very. popular film study also on the we see the be the weekend do math they can get one so they can have what you should and substructure of with one million euros one have to say that the forecourt use it's also things are just hard core the there. their instincts also things and there is a park which which rooms. yet so it in goals for here you know the answer for example have some kind of collective intelligence each one is maybe not the is bright but together we they can deal thinks that they can act as intelligent sold examples of natural intelligence i would hear some warm us and and he says the elegance the four. example u.k.'s exactly twenty euros to think and the two hundred eighty two from things like the show. when and and here is that this form just want a dozen everything your arms but is an animal rights or you most vulnerable in all doesn't it. and. white oak able to the so obviously the number of your own somehow correspond to the diligence some are is not the only condition but it is it is it part of mechanics behind the intelligence would how it works in the people who do the deduction servants know a little bit how it works but here's an example. apple is one year old. and in your own harris the has a soul my here and he says the woods kind of decision maker it's here then it here's the deal readers can reduce our the sensors with that sensors or source they never felt. and then and this all modest sites if. the process some signal through the extra or not and the exxon is the ultimate action of the euro and goals then moved to mosquitoes or two other neurons and the gap and actually the what happens here it's very interesting rights of how all this sell one still the scientific. the signal for or not. miss. i. and the white why i would hear these this elegance think it's very interesting because this is the only a small that that we have a chronic films all the conventional full neurons of it there are not so many right and it's constantly updated but there is the reason the or for the architecture of this war that say we. will. and if we if we go a little bit back to the to the apes then an interesting one reason may be why the where different to the apes is that we can speak so the language was very important ability that gave intelligence support of intelligence. and the apes they can have physically they can all see if they the the filthy learn to the they think they can learn to read or to communicate they can speak about future and past they can speak about sign language they can speak about the. epsom people that are not in the room they can tell stories but they cannot actually speak and the these this was one of the main reason why the didn't may be right the deal infusion. would have. its chief. if the exactly and then you're on the within the year on the signal that is the transporter are within your means these are particles electrically charged particles and if you can you can change dramatically some matt mice for example such that if its signal comes to the accident a year on. the lights up then you can put some microscope and the actually have sort of which parts of earth the brain the r.f.u. for specific actions and how it works. roots. the and now it's that's a little bit artificial intelligence what we have now.
and yet it's just just syrup simple of what we have no off or would you think that this is what are we are right then this is the in the car an automatic the braking system and the disease. it's become clear enough. we think that the water border is somewhat efficient engines. is it into agent. no but it also went to switch off the water at nos one seizure of it is it is a of course it does the most read but the invisible based system it's hard for that it's likely be this thing is cut for that so if it shows if you would the hoop hear a market instead of the boiler. market which would just switch off the water. there may be people who think it's it's like intelligent right now. food. when does a system him artificial intelligence the with the system should the other thing quartet and it will do in two ways like a human does oration on and of the human is not always original because humans have emotions or they own the river. and says and they do not actually very seldom expedition more like a human for the same thing or act at the end these these up for pars thinking and acting like a wonder is a the science of cognitive science topix a you can study it. there was a psychology role model in this area we will see how machine can think an exploration of the. and there are to be the parts in the artificial antigens actually won the systems that are only based where you can put a lot of motion the system and and it can help you and one which the can learn from its environment and this is exactly the machine for example through based systems to use for example a program.
reports that you actually put the rules in informal support your putting something and then the machine just for the rooms except it's a cute some are some other reasons and they're also knowledge based system used for example in the images seen or other areas which help human to. find the right to make the right decision so you just a number of facts or which in other human action record and in broughty and the their knowledge of europe's here with the you probably will work a lot on the yacht with she's just a little bit different organisation often wallet. which and the but still it is it is a database of of facts or relations. from these systems were very very popular in the past where the machines were no not really perform on and the one nevertheless we wanted to let them do as much as possible so we put as much knowledge as possible into the machines are but soon you will see that if you for example. people for for the medical them in a four for any other the mean you will you will notice that you can have everything we always knew facts. it's a lot of facts if you do the reasoning its it just takes a lot of time and the it's why the other approach for the good people of haiti with the with the better machines with the more data and is a machine learning and the the idea that machine learning is that the application. and can improve their performance through governing experience seeing the more they take samples and the. we have some some measurements the formally that the computer program is on from experience the with respect to some loss of tasks and therefore months there are so we need also to measure the to have to say culpable if these if he gets better and better over time to sort them. examples of machine learning are are here the computer games are right you can play computer the game against a computer and if we learn from from human and getting better and better you can live machines play against each other from. and the or you can you can use lots of examples of letters year and the band the machine will be able to recognize written character and met them fool of. characters. and he may be interesting to say here that the machine i mean we are learning the human being a learning the the they can we can learn up to one hundred percent accuracy one task but insects for example they cannot they always the number one hundred percent there's a trade will be bitter. in its exploration and exploitation the a we are ok so the they will never be able to go on for sense of the assumption that the machines only works if it's very important the data sets for training in his to be representative for the problem and it has to. for all the aspects of the problem and properties all the training consensus should be relevant to the problem even if these assumptions if the street victors for the couple this think that there's then the more those approved the get from from from the thinning are also useful otherwise their biased i will not grow. to bias in the stock because the professor martin who will work of exactly about the bias all about the data which the the the can be used for training camp in with other problems is that these not not good. there are are three the for. machine learning the eerie us so one his answer was snoring. a super was lord and important warning in the difference is not the not always in the output on not mean that all pulled off of all the algorithms but in the way in the procedure how we learn it's super was warning means the during the only procedure you can tell the machine if it's going right or wrong during the procedure. with answer but i don't think you just give the machine the data in the figures out itself. without any particular goals and the it looks for example for the pattern in the data and so on and reinforcement warning. is a the only procedure a course without. without the jets are just one so you don't tell if it's good or not but you look the results and the punishing the other even if the results was wrong or the water the algorithm if the result was good so unfortunately owning the will try different possibilities to was also a problem and the the. it will stick to those which which were a bar that reward that the most but we will not the it will be the difference today. how much money can be useful for for different purposes for example for. the you can use it to learn the normal behavior or specific specific data streams and detect outliers for example as effects for some events in the industry which the role of the union even detection movie also one task in these the swing for. don't ask in this project. there are over the filtering for the commander systems. you therefore suffocation algorithms for example to detect objects on the images of and eric will tell about is also later and the there are also costing and classification of the reasons of working together for example to extract popular topics aura again to detect dick even. and the popular topics and new topics which of years so with the. that even more ball before. this photo was in a very simple example and i have here my might mime. the agent let's say and i want to make intelligent i would like to make an expert on the car market. it should go to the to the web page to look at all the cars that are present that and became an expert and helped me to make decisions ok of course machine can not the work with the with objects it works with measurements so what i should do first i should define the properties of these objects. and the somehow with them with him in in in such a table rights aware. for a game where were active roles where you troll news is it is an object. and they have also the table where each call is a specific property or feature so if you hear proper to feature if it's like similar things i have a brand year for these guards point slide and built in germany intellectual and the the eighty two and so on. now if i have another car and i also extract the properties of this car and put this in the table are suddenly becomes possible for the machine to compare the school objects and people just a big one phone for example the year and the compare.
the day and compare to compare the year so we can say here that one carries older than another one right in one dimension but we could also phones also dimensions of course we cannot put it in the of more than three dimensions the and also more than two.
you will for musician was this little more do with just will have to saw this is a thousand kilometer so-called whole or how much of the it was running already on the street. if i end this these two features actually the spanish victor space and each object become a particular point in these sectors space like like these two cars and now i'm additionally the able to move to use see the distance between the car so most. i don't have just the measurements just the future but also can have a friend of for similarity measured between the car and my the my hope is that if i. take a lot of objects with their properties and and put them into into these victims face in the senior or projects will be at the same location in newspapers things like you're older folks while and hear all the media or. and the i can also hear some smaller here and these victims face finally it helps me to describe my data set to compare objects the rank the objects maybe and put them in groups. now i assuming that the my my algorithm doesn't know anything about the label of the data so whether it's the sports live in or near or something else he just sees the the features year and the out the thousand kilometers from my whole police. that he will be able to find find groups in the in this rule if the it's visible from the distance and the the skin look like these these these approaches cold blustery and the i can do it than on the data set without a specific goal i just feel he hears a. the of what is the structure of these the to the species around of it. so i can additionally detect better and detect also on the market is i can also detect year that the soldier is somehow different from the wrist but not the old the time i have these these nice distance between two clusters of with into groups most of the time. and it looks like nice and here i can use the. you're sick and here i can use a sound of a super was ordering were ideal the computer holocaust believe these the displaced in the into groups so actually a tale which which which neighbors are if we object and the task of the other. it is now to separate these to distribution is so to find some kind of separating line separating border hear from these to objects and the this is this is my mother then at the end i can use it to decide if i have a you'll get somewhere here i can decide of a disease looks like on the rights. but it's the same your or its on another site if the folks won. so this this this is sick of it basically about the basics are off machine learning curve now i have some slides of all the room for the development.
the or or or the application in in different areas and here's a slight the ball the performance of the the euro much recognition from two thousand there until two thousand and seventeen. the and the weekend problem rather simple minutes until two thousand and twelve and. and the error here although there is a competition the mission at largest usual recognition competition where you have a large data sets with one thousand classes and the you can create a model classified as dataset. the competition between different models and two thousand and ten for example the air or or do we not hit the air or zero those who ate. one percent so this is the percentage of the images which were only classified and the two thousand twelfth i think here the the new owner for skin right and the air or their own went down when down and in between two thousand and fourteen and two thousand and fifteen the air or was this. then the error of leverage anyone who'll an updated is davis and the i think they the media run these competitions two years more there or was then sold little it and that they decided to stop this competition doesn't make any sense and the loop for more interesting problems. and one of these problems of course the chicago edition which is not solved yet the really for everything for but the similar competition or seem to have seemed only their there is there is a pop five need to sit and where you can test your morals the but i think it's.
this particular english and and in two thousand seventeen also the first models for bt the day the the average human so the the mortars nicholas air or discussion than a human but again so it was a particular english language she doesn't work for all the language you see it. but the you see that the air or falls down. and the in machine translations these are ok for from a disabled by the image no here on the y. axis we see some score for machine translation so that the higher the better and on the on the x. axis there are different works in the torture which which war there. executed on on the data set and the action is week we can see we are far as of the deadline is the human performance the human reach this course of fifty and the mothers are not so far but the is one can see the it's his is going better and within the and. i'm here three years in almost every should this source to these line so in couple of years might also have break these line and we can hear from the machine translations. and here again the this is a these also english french or a specific language film or more for for everything but the progress is is very impressive. but at the it the line. and here the overall much and on a course occasion example of education example if so you cover all data you extract some features out of these data and to use this features with some other than to train a model and this one though it helps you to decide for the objects which you never saw before but you know the future to decide. to to doing for the missing features like your of the that the price of these is basically the complete moral and the also some role in both the so did the industry the heart of our meters described the configuration of the model they even by users so for example you would like to. see there are three he clusters and find new three courses in the desert and their party leader in for a metre i learnt his part of the model and their learned by the more.
flute or no maybe some questions before we go to the un super worst learning part. but. no.
so the answer was soaring the boston has already said that we can use to describe the data to partition the data to find some patterns and detect anomalies. and here i was community in the office called before it's just a web page where you can sell your car and i created former said his dance. to two hundred and the bloated hear the price on the x. axis and the the year on the on the rest of the y. axis and the year on the x. axis and the i was wondering that the the prize in the year somehow the are not correlating right. any idea what skipping cure why use looking like these and he was looking like these each point is a particular car from the data set. five. i. to mean what it's a possible future. but also of all time or six episode is that these are two different distributions indeed these a just two different distribution also the labour is actually the same. one example here you see that the the new of the car the more it costs and he is also forcing right and the knowledge i would like to have another it on which which finds me these different distribution so just give a date and and it should tell me which differ industry which is i have my data said. them in these is that your partitioning i the. i want to partition database in the costs and every object in this particular case should belong to only one class others and within one class that the object should be similar and be if you take the objects from two different lost jobs they should be his dissimilar as possible and the possible solution would be i. i just read the for a possible partitioning call the data said. and the and then the somehow with semitic will identify the best one but this will of course they will do with a brute force but these rules of course they are lots of them will suffer for. it's why where one to use your he sticks for partition the space years one overview where every koran is actually different are reading them and every role here is a different data set and to see that you yeah there is no off the ball rather than. actually for all data sets but they will behave differently and the these is the first problem how to select the right the right i'll get in for my problem and machine learning what you do you do a lot of just trial and error and the but once you're going to fight moralists with model for for your problem you can cause use it again and again. and. your eye. the organ. than maybe maybe maybe one what issue before so this is a what what what we are going to see is the canes are very than anyone not familiar with teens clustering. everyone familiar. no then we have to have it can be very very simple logo earth and and thirty the scan so density base clustering anyone more familiar from the technical people so everyone familiar with the new we've all very fast after the coffee break. it was ok. with.