Deep Learning with Tensorflow - Autoencoder Structure Video Lecture | Tensorflow: Learning made Easy - AI & ML

hello and welcome in this video we will
provide an overview of the key
components of an autoencoder
as well as the applications that an
autoencoder is suited for an autoencoder
also known as an auto associate or
Diablo network is an artificial neural
network used to recreate a given input
it takes a set of unlabeled inputs and
after encoding them it tries to extract
the most valuable information auto
encoders are used for feature extraction
learning generative models of data
dimensionality reduction and even
compression a 2006 paper showed better
results than years of refining other
types of networks so this was a
breakthrough in the field of neural
networks which was stagnant for ten
years
Auto encoders are based on restricted
Boltzmann machines and they were
employed in some of the largest deep
learning applications they are also used
as the building blocks of deep belief
networks as an example let's say that
you want to extract the feeling or
emotion of a person in a photograph this
is a small image that's just 256 by 256
pixels but this means that the dimension
of the input is over 65,000 as we
increase the dimensionality the time to
Train and fit our neural network
increases exponentially so we need a way
to extract the most important features
of a face an auto encoder works well for
this type of problem an auto encoder can
be divided into two parts the encoder
and the decoder the encoder needs to
compress the representation of an input
in this case we're going to compress the
face of our actor moving from two
thousand dimensional data to only thirty
dimensions the decoder is a reflection
of the encoder Network it works to
recreate the input as accurately as it
can it has an important role during
training
and that is to force the auto-encoder to
select the most important features in
the compressed representation after the
training is complete you can use the
encoded data that has been dimensionally
reduced for the application of your
choosing this image compares a two
dimensional reduction for 500 data
points of M NIST a database of
handwritten digits the left plot was
generated by principal component
analysis and the right plot was
generated by an auto encoder you can see
that the auto encoder provided us with a
better separation of data an auto
encoder uses a loss function to properly
train the network it then uses gradient
descent to reach the local minimum of
the function you can use different
functions for binary values and real
values please review this notebook for a
more in-depth explanation of these
formulas by now you should understand
the basic structure of an auto encoder
as well as the benefits that an auto
encoder provides to different
applications thank you for watching this
video to practice and learn more go to
the lab and run the code for yourself