►
From YouTube: Automate your life with Azure - Gergana Young
Description
Machine learning, neural networks and image processing are things that require quite a bit of in-depth theoretical knowledge to implement. They have so many every-day, useful applications that we take for granted. Facial and voice recognition, natural language processing, even something as simple as auto-complete, all use some form of machine learning. Surely, there must be a way for us to use the power of machine learning without having all of that algorithmic knowledge. The short answer is; yes, we can.
For the longer answer, and to find out how, join me for this talk. Using the cloud (the Microsoft one in this case) I will show how easy it is to add facial recognition to your application, and automate some of the simple tasks we do every day. No mathematics needed!
This talk was recorded at the Jozi.JS meetup group on 22 November 2018.
https://www.meetup.com/Jozi-JS/events/254136610/
Disclaimer: The opinions in this talk are those of the individual and do not necessarily represent those of this channel, the meetup group, or it’s sponsors.
A
Hello,
everybody,
its
Mike
said
my
name
is
Jerry.
Well,
actually
my
name
is
Gary
Anna,
but
I
go
by
Jerry,
because
that's
a
little
bit
easier
to
pronounce
and
today
I'm
gonna
talk
to
you
about
automating
your
life
with
Asher.
Oh,
not
really!
Your
life,
not
your
whole
life,
a
very,
very
small
part
of
your
life,
but
before
I
get
to
that
I
need
to
tell
you
a
little
bit
about
my
team,
because
this
whole
talk
revolves
around
my
team
and
what
I
did
for
us.
So
if
this
thing
was
on
this
is
my
team.
A
But
so
the
team
consists
of
me
already
introduced
myself.
Then
this
Rory
Rory's
not
a
horse,
he's
a
dwarf,
but
he
thinks
it's
funny
that
he
has
a
dwarf
pony
as
his
picture
on
Twitter
then
is
Tony.
Tony
doesn't
speak
that
much
in
community
yet
we're
getting
there,
though,
and
then
there's
Mike
person
who
introduced
me
to
those
of
you
who
don't
know
him
our
local
JavaScript
expert.
A
So
we
started
with
automating
a
lot
of
those
tasks
on
trailer
and
we
use
something
called
Butler
on
trailer
for
automating
those
tasks
and
it's
really
good.
It
works
really
well.
It
reminds
us
when
we're
missing
out
on
those
tasks
that
we
forget
to
do,
but
I
wanted
to
take
this
further
I
wanted
to
make
it
so
that
when
we
have
our
stand-up,
it's
a
lot
easier
for
us
to
do
so.
When
someone
stands
in
front
of
the
computer,
when
we're
doing
stand-up,
the
Trello
board
should
be
able
to
filter.
Who
does
that
present
tasks?
A
A
A
But
now,
as
I
was
speaking
to
a
PhD
student
who
is
doing
machine
learning,
he
was
telling
him
this
whole
thing
about
how
I'm
brought
in
your
own
network
once
and
he
said
to
me,
but
was
that
before
2012?
That's
it?
Oh!
Yes,
it
was
before
2012
and
he
said.
Oh
then
it
doesn't
count
so
I'm
not
really
sure
what
exactly
happened
to
neural
networks
in
2012,
but
apparently
something
big
happened.
A
So
now,
I'm
standing
here
telling
you
that
I
don't
know
much
about
neural
networks
or
machine
learning,
but
that
we
require
this
in
order
to
do
facial
recognition.
So
how
are
we
going
to
do
it?
Well,
what's
the
cloud
the
Microsoft
one
to
be
exact?
Okay,
so
just
cognitive
services
provide
a
bunch
of
different
services
that
allow
you
to
do
these
things
easier.
A
A
I
was
female,
set
out,
there's
100
percent
chance
that
I
am
smiling
in
that
photo
said
I'm
not
wearing
glasses,
and
it
said
that
I'm
27
years
old
now.
This
last
part
really
amazed
me
because
in
that
photo
I
am
27
years
old.
Now
this
was
amazing.
I
thought
that
was
your
computer
vision
was
the
best
thing
ever
because
even
a
human
looking
at
another
person
can
tell
exactly
how
old
they
are
right.
A
A
A
Now
for
those
of
you
that
know
Rory,
you
know
he's
not
50
and
with
most
people,
when
you
tell
them
they're
older
than
they
actually
are,
they
kind
of
get
offended.
So
I
said
to
him:
hey
Rory!
Is
it
okay?
If
I
share
this
with
people
like
I'm
gonna?
Do
this
to
a
talk
was
in
a
talk
to
lots
of
people?
Is
that?
Okay,
if
I
tell
them
that,
as
you
said,
you
were
50,
he
said
it's
not
a
problem,
and
now
this
is
a
quote.
A
A
Oby,
two
years
of
marriage
had
aged
me
by
10
years,
Oh,
see
three
months
of
being
on
this
team
had
aged
me
by
eight
years,
either
way
I
thought
this
was.
This
was
great.
It
was
working.
It
was
doing
what
I
needed
it
to
do
so.
I
moved
on
and
I
started
hacking
away.
Now
there
are
two
parts
to
facial
recognition.
A
Firstly,
identifying
where
the
face
in
the
photo
is,
and
then
saying
whether
that
face
is
that
person,
but
in
order
to
be
able
to
give
and
sign
a
name
to
a
face,
you
need
to
train
it
with
pictures
of
that
face
now.
For
me,
I'm
a
millennial
I
have
tons
of
photos
of
myself.
However,
the
rest
of
my
team
all
have
kids.
Now,
who
knows
what,
for
those
people
with
kids
have
any
one
photos
of
the
kids?
A
Exactly
yes,
not
photos
of
themselves
photos
of
the
kids,
which
meant
that
getting
between
three
to
ten
photos
of
my
team
members
was
not
very
easy.
It
did
involve
a
bunch
of
trips
to
our
marketing
team
to
ask
them
if
they
had
photos
from
previous
BBD
events
that
I
could
use,
or
there
was
a
bunch
of
times
when
I
would
be
like
hey
Mike
look
up,
I
need
a
photo
of
you,
and
then
there
was
a
little
bit
of
social
media
stalking,
but
eventually
I
got
the
photos.
A
I
needed
and
I
started
training
my
system,
so
the
way
it
works
is
the
cognitive
services
have
a
whole
bunch
of
sections
to
them.
There's
vision,
speech,
language,
knowledge
and
search.
The
part
we
we
are
using
is
the
vision
part
and
the
vision.
One
is
also
separated
into
different
things.
So
you
can
do
things
like
identifying
text
in
an
image
as
well
or
using
Azure,
but
the
part
we
need
to
use
is
the
face
part
and
two
sections
of
that
face
part:
the
face
detection
and
the
face
identification.
A
What's
really
cool
about,
it
is
if
you're,
first
starting
out
with
this,
you
get
a
30
day,
subscription
key
that
you
can
use
and
you
don't
need
to
even
like
register
and
have
a
Microsoft
account
or
anything
like
that.
So
if
you
just
want
to
try
it
out
and
see
if
it's
gonna
work
for
what
you
needed
to
do,
you
can
just
go
to
this
website
and
do
the
trifle
free
on
any
of
those
and
you
get
30
days
for
all
of
them.
A
So
you
could
do
all
of
them,
but
first
part
was
to
do
the
setup
and
the
training
and
it's
a
restful
service.
So
the
first
thing
we
need
to
do
is
create
a
person
group.
Now
this
person
group
has
an
ID,
which
is
case
sensitive,
that's
important
to
remember
as
well,
because
otherwise
you
end
up
creating
multiple
person
groups
and
you
end
up
confusing
yourself
so
case.
A
Once
you've
created
the
person,
you
associate
photos
with
that
person
who
add
faces
as
they
call
it
and
the
add
face
works
by
you
upload
a
photo-
and
you
say
this
is
this
person
right,
so
it
first
runs
the
detect
to
find
the
face
in
the
in
the
image,
and
then
you
say:
okay,
this
is
this
person.
What's
important
here.
Is
that
that
method
has
to
have
photos
where
there's
only
one
face
in
that
photo.
A
So
there
has
to
be
only
one
person
can't
be
other
people,
okay,
once
you've
added
all
of
these
faces
and
associated
them
with
the
people.
You
core
train
method.
This
train
method
runs
a
hope
for
not
a
very
long
time.
I
mean
for
the
four
of
us.
It
ran
for
about
less
than
a
minute
with
ten
photos
each.
So
it's
very
quick.
The
only
problem
with
the
Train
method
is
it's
asynchronous,
so
you
can't
really
tell
when
it's
done
so
you
have
to
keep
calling
the
get
training
status
method
until
it's
done
to
see.
A
So
you
create
some
sort
of
timeout
to
call
the
get
training
status
every
so
often
to
see
if
it's
done.
Training
once
you've
trained
everything
we
need
to
detect
and
identify
so,
firstly
face
detect,
so
you
upload
an
image
and
it
detects
all
of
the
faces
in
that
image.
It
can
be
multiple
faces
in
this
case,
and
then
you
call
identify
on
all
of
those
faces
now.
A
This
is
the
part
where
I
struggled
a
little
bit,
because
I
decided
to
do
this
in
angular
and
rxjs
is
a
great
library,
especially
when
you
know
very
well
how
to
use
it
and
you're
using
it
to
its
full
advantages
when
you're
using
it
just
because
you
need
to
call
a
bunch
of
rest
rest
services,
one
after
the
other,
it
gets
very
confusing.
There
was
a
lot
of
merge
mapping
and
folk
joining
and
a
lot
of
results
that
I
didn't
even
I,
didn't
even
care
about
the
results.
A
I
just
needed
it
to
do
something
so
that
got
a
little
bit
confusing.
But
after
a
while
I
figured
it
out,
I'm
not
actually
going
to
show
you
that
code,
because
it's
not
the
best
code
I've
ever
written.
It's
a
very,
very
long
method
of
merge,
Maps
and
the
other
thing
that
I
had
a
little
bit
of
trouble
with
here
is
that
when
you
call
the
detect
method,
you
need
to
send
binary
data.
A
So
you
need
to
change
your
header
to
be
to
your
request,
header,
to
have
an
octet
string
as
a
type,
whereas
when
you
do
the
identify
method,
you're
sending
JSON
data
because
you're
just
sending
an
array
of
faces,
and
then
you
have
to
change
it
back
to
be
jason
and
I
kept
switching
between
them
and
I
kept
switching
in
the
wrong
places.
So
it
took
me
a
while
to
get
that
right
as
well,
but
I
got
it
right
eventually.
A
No.
The
last
time
I
did
this
talk?
Okay,
so
I've
currently
trained
my
neural
network
on
mic,
Tony
and
Rory,
but
not
on
me.
So
you
can
see
there.
It
says
that
my
face
is
unknown
right.
So
what
we're
gonna
do
first
thing
we're
going
to
do.
Let
me
just
make
sure
that
this
is
still
running.
Okay,
it
is
store
any
is
we're
going
to
get
some
training
data.
A
It
should
detect
my
face
as
well
and
the
way
the
trailer
part
works
and
what,
as
I
still
haven't
gotten
it
to
do
it
automatically
so
open
a
trailer
in
a
new
tab,
but
for
now
it
just
works
where,
if
you
hover
over
the
person,
it
gives
you
a
trailer
link
at
the
bottom
to
that
person.
Okay-
and
you
can
see
it
still
does
all
of
the
attributes.
So
the
age
gender
smiling
glosses.
It
still
does
all
of
that
stuff,
because
that's
what
the
face
detectors
and
there
are
other
attributes.
A
So
those
are
the
four
attributes.
I
chose
to
read
out
of
the
facial
recognition,
but
there's
other
things
like
facial,
hair
and
emotion,
being
displayed
and
a
whole
bunch
of
other
stuff
which
you
can
play
with,
which
is
pretty
cool,
so
I'm
gonna
upload
one
more
photo
one
with
Rory
in
it.
So
you
can
see
that
that
works
as
well,
and
it's
loading
there
we
go
so
you
can
see
there,
we
go.
It
is
identified
Rory
as
well.
A
Now
one
of
the
interesting
things
here,
if
you
take
a
look
at
mine
and
Mike's
faces,
are
in
orange,
while
Rory's
and
Tony's
are
in
green.
That
is
because
it
also
gives
you
a
confidence
rating
of
percentage.
How
confident
it
is
that
that
is
that
person
so
well
have
what
I've
made
it
do
just
for
to
make
it
a
little
bit.
Prettier
is
if
it's
over
80%
its
green.
A
Right
so
what
I
achieved
with
this
research?
It's
pretty
much
learning
that,
no
matter
how
complex
something
may
seem,
we
as
developers
can
do
it,
no
matter
how
complex
that
calculation
looked
I
mean
I,
have
no
idea
what
that
calculation
said
right.
It
seems
like
magic,
but
any
sufficiently
advanced
technology
does
so
that's
pretty
much
it.
Thank
you
very
much.