What is Unsupervised Learning? Matthew Voke English English (auto-generated)

What is Unsupervised Learning?YouTube↗ IA↗(long shot)
Education 59 views 6 likes 0 dislikes 03 Apr 2020

Matthew Voke 1.8K subscribers United States YouTube↗

Click to Play Video

What's up YouTube Matt Voke, Callsign Jerry, with another bite-sized AI video

what's up YouTube Matt vote callsign Gerry with another bite-sized AI video

last video we looked at supervised learning holding AI's hand and

last video we looked at supervised learning holding ie eyes hand and

helping it determine correlations in data

helping it determine correlations in

what's the next logical training method? unsupervised learning

data what's the next logical training method unsupervised learning

if supervised learning trains artificial intelligence with label data

if supervised learning trains artificial

unsupervised learning obviously trains AI without labels

intelligence with label data unsupervised learning obviously trains at AI without labels

where supervised learning requires that

where supervised learning requires that

significant upfront cost in order to label massive amounts of data

significant upfront cost in order to

unsupervised learning doesn't require as significant of that initial lift

label massive amounts of data unsupervised learning doesn't require as significant of that initial lift it is

it is possible without hand-holding from humans

possible without hand-holding from

the downside is a lack of guidance and typically more significant data and

humans the downside is a lack of guidance and typically more significant data and

computational requirements to teach those models with less predictable

computational requirements to teach those models with less predictable

results so what's the use of unsupervised learning if it requires

results so what's the use of unsupervised learning if it requires

more data and compute what if we don't know how to label the data many times we

more data and compute what if we don't know how to label the data many times we

don't even know what patterns or correlations actually exist

don't even know what patterns or correlations actually exist what if we

what if we don't want to hamstring our discoveries of insights with human imprinted

don't want to hamstring our discoveries of insights with human imprinted

heuristics bias preconceived notions or what correlations should exist

heuristics bias preconceived notions or what correlations should exist well then

well then you have unsupervised learning my friend

you have unsupervised learning my friend

the most common and perhaps the easiest

the most common and perhaps the easiest

unsupervised learning method explained is clustered detection

unsupervised learning method explained is clustered detection we can easily

we can easily demonstrate this on XY scatterplot

demonstrate this on XY scatterplot

although there can be significantly more

although there can be significantly more

dimensions than this 2-D example here we can semi randomly plot dots on an XY

dimensions than this 2d example here we can semi randomly plot dots on an XY

scale some of you may already be able to see that these dots aren't really random

scale some of you may already be able to see that these dots aren't really random

but you can't really make too much out of it unsupervised learning is able to

but you can't really make too much out of it unsupervised learning is able to

help us by suggesting that we break these dots up into three categories or

help us by suggesting that we break these dots up into three categories or

clusters the AI is able to draw the most efficient cluster around these data for

clusters the AI is able to draw the most efficient cluster around these data for

us now imagine these dots aren't just blue dots their height and weight their

us now imagine these dots aren't just blue dots their height and weight their

length and color their price and size any feature you can think of the key is

length and color their price and size any feature you can think of the key is

that we don't even need to know these features we don't really care we let the

that we don't even need to know these features we don't really care we let the

AI solve the problem while we're blind at least initially that's pretty

AI solve the problem while we're blind at least initially that's pretty

powerful we can potentially solve problems not even knowing what the

powerful we can potentially solve problems not even knowing what the

problems are taking this one step further, a couple of years ago Stanford and

problems are taking this one step a couple of years ago Stanford and

Google researchers applied these techniques to unsupervised learning on

Google researchers applied these techniques to unsupervised learning on

YouTube and interesting things happen they were able to build high-level class-specific

YouTube and interesting things happen they were able to build high-level class

feature detectors from only unlabeled data in other words they

specific feature detectors from only unlabeled data in other words they

discovered the features of cats with an AI just watching YouTube. Big whoop right?

discovered the features of cats with an AI just watching YouTube big whoop right

Wrong! double rainbow all the way across the sky. The simple fact that he was

wrong double rainbow all the way across this yeah the simple fact that he was

able to learn these high-level features of both cats and humans from just data

able to learn these high-level features of both cats and humans from just data

tells us a couple things 1) there's a heck of a lot of pictures of cats on

tells us a couple things one there's a heck of a lot of pictures of cats on

internet 2) this was the first time that anybody showed that it was possible to

internet - this was the first time that anybody showed that it was possible to

train neurons to select high-level concepts like this without labeled data

train neurons to select high-level concepts like this without labeled data

they literally just let it go and it figured out these features on its own

they literally just let it go and it figured out these features on its own

number 3) AI is pretty dang cool.

number 3 AI is pretty dang cool so

So, unsupervised learning trains AI from

unsupervised learning trains AI from

data without labels while supervised learning requires a significant upfront

data without labels while supervised learning requires a significant upfront

cost in order to label massive amounts of data unsupervised learning does not

cost nor to label massive amounts of data unsupervised learning does not

require this initial lift but it typically needs more data and compute

require this initial lift but it typically needs more data and compute

huh weird data is king!

huh weird data is king remember to smash

Remember to smash the thumbs up, subscribe if you want to

the thumbs up subscribe if you want to

hear more of this beautiful voice and isn't AI fascinating for crying out loud

hear more of this beautiful voice and isn't AI fascinating for crying out loud

it taught itself cats

it taught itself cats I'm sorry I'm

I'm sorry I'm thinking about cats again

thinking about cats again thanks for

thanks for watching keep learning this is Jerry, SEEYA!

watching keep learning this is Gerry see

Similar videos:

Unsupervised Learning
Introduction to Data Analytics
10.2K views 15 likes 3 dislikes 21 Jul 2015100% match

Unsupervised Learning
Udacity
21.6K views 28 likes 4 dislikes 23 Feb 2015100% match

Unsupervised Machine Learning
The Foundations of Biomedical Data Science
710 views 7 likes 0 dislikes 24 Feb 201767% match

Unit 6 1 Unsupervised Learning

7.4K views 3 likes 0 dislikes 22 Oct 201167% match

Settings Privacy About Contact Donate