PyTorch Lecture 01 Overview Video Lecture - Pytorch A Complete Guide -

welcome to mal TL for everyone with a PI
torch my name is Sam Kim from HK SD this
lecture one will offer up your entire
course what are the goals of this course
so we want to provide basic
understandings of machine learning and
deep learning but not only the theory
part we are going to implement Everett
concepts using Python so that we can
have concrete understanding of these
concepts in order to enjoy this course
we just need basic math skills including
algebra and probability also you have a
some experience of writing software
using Python there will be enough I
assume that you are standing Samaritan
here so hopefully our lecturer can help
you to move towards for the experts so
that once you have any problem solve you
can use machine learning and deep
learning skills what is machine learning
before we talk about machine learning
let us think about what is human
intelligence using our intelligence we
make a lot of decisions but how we gonna
do this
for example lunch what do you want to
eat for lunch
how we can make this decision so usually
how we do is that we get some
information like history confirmation
what did we like what we didn't like
using all this information we can try to
decide what to eat for lunch so for
example here let's eat pizza right so
this is typical way to make a decision
and then we usually saying this one as
inferring using some information
how about what address so same so you're
gonna use certain information based on
information we think and they read the
side also we can provide some in some
images like this and the v-neck we can
ask what's inside of it this picture and
then using this image information as
input we can say oh it's CAD
so this process can be prediction or
inferring also this problem for example
we want to know that the final grade if
the students spent 4 hours of studying
so if you have some knowledge about the
colors or some histories about this
course probably we can answer these kind
of questions this is our human
intelligence the machine learning is
exactly the same thing so we want to
exact the same thing but using machines
that is the machine learning for example
here while to dress we're gonna provide
a lot of information about it and then
this machine learning can predict or
import the output so image the same
thing so we can provide the image
information the machine learning can
tell us what is the image apart so this
is machine learning and then this number
the same manner we're gonna provide this
information to this machine and this
machine can predict what is number in
this image also we're gonna provide the
information about how many hours they
studied and that this machine can tell
us what will be the points they might
get so this is one example of machine
learning and then let's just think about
this problem a little bit so we're gonna
feed all these pixel information to the
machine then how they gonna know what is
number inside using the machines are
really stupid so they don't have any
idea what is this pixel pixel means or
what is even number two or means so we
need to teach these machines so in this
process is called the training you know
train machine what we need is the width
of images
and then corresponding true labels so
this one you usually call the label the
data set we're gonna provide a lot of
data sets and then we train our model
once it's a trained now we can ask by
providing some image information ask
what is number inside that's why we
called user this is machine learning but
because we teach this machine to a
certain job
similarly this problem what we're gonna
do is that we need to provide some sort
of historical information for them
previously one hour
they studied one hours and then take a
two points two hours four points so
using this information which train our
model and this one is called the
training data sets and then once the
model is trained we can ask this model
machine owner that for four hours what
will be the points they may get this is
the typical machine learning concept and
there was a difference between AI
machine learning and deep learning so
I'm gonna use diagram from this book
then AI in general is anything that has
some sort of intelligence then we call
that as AI in order to help that we need
we need some solutions there are
actually many solutions one solution
where one approach is called machine
learning and we discussed we provide
labeled data set with China model the
ones a trained we can ask some questions
and then inside the machine learning
there are many many algorithms of course
and then deep learning is a group of
algorithms that are using neural nets
and then usually the neural nets are
really tipped many many layers so we
call that as deep learning
while we care about it learning so one
simple answer for that is that deep
learning is currently working very well
so I just have provided a few examples
in this slide so one example is the
video location answering so we provide
the one picture and an desk
all kind of questions about this picture
and that this neural net can provide the
reasonable answers also we can provide
the certain images and then we ask a
neural net to change the style of these
images and then you can change the cells
amazingly very well it's called a style
transfer and we have many many many more
amazing examples you can find more demos
in here why we care as a developer so in
fact we can write the software without
even knowing deep learning or using deep
learning so this one is called usually
rule-based to systems so we can imagine
all possible inputs and then based on
the inputs we design all rules manually
and then hopefully these rules can work
with these inputs and then generate some
useful output machine learning can
change this one significantly instead of
making all these laws manually this
machine learning can learn these rules
automatically but still one limitation
we need to provide very nice input for
the machinery which is called features
so usually we design these features
manually but what Stepfather the tip
learning can change this one completely
so in these cases we call the
representation learning we don't have to
select what is good feature not good
feature just to provide all possible
inputs and them in the multiple layers
this tip learning can learn its
representations automatically and then
figure out what is the kilt
good picture what is not good picture
and the news good pictures and then
produce
reasonably could have
Kono's is even though the still burning
sounds a little bit complicated but it
is very simple to study EPA basic math
skills with a programming experience you
can catch a most of these concepts
within one year even in six months
also there are many of frameworks that
you can try to use and you can
understand them more in the internet the
lots of study resources including this
lecture and most of all it's really
really fun if you just start studying
you cannot really stop in our lecture we
are going to use Python to implement all
the basic concepts so pi touch is
framework who implemented tip neural
network why PI torch so we selected the
Python for these two regions first of
all it's more pythonic which means that
it's flexible it's intuitive and then
it's very easy to debug so second reason
is that it's more neural networking
means that you can just imagine any
neural net and then you can just put
these in your imaginations to the code
and that's it you don't have to learn
any new schemes or ways to implement new
net it's very natural to implement your
neural net directly so this is a
hands-on icon in our slide so let's move
on to our hands-on section so what we
want to do is that we want to install
Python we can pease pottage that work
and then based on your operating system
you can get some instructions how to
install so if I using windows also there
are many ways to install Windows once we
figure out you can test so this is your
exercise 1.1 if you haven't installed
the PI touch maybe after this lecture
you can try to install one in your
computer apt install installation you
can test it python 3 and then you can
try to import the torch and then we can
print out this and they will get
the the right person number something
like then we'll be very happy the rest
of the course we're gonna cover some
basic concepts linear logistics of the
mix models and then also we're gonna
cover DNA and CN n + RN n which are
mostly important tip neural net
architectures and they were gonna write
everything in Python next we're gonna
talk about linear models