Press "Enter" to skip to content

How I Searched My Way To A Cure | Abu Qader | TEDxTeen


[Music]
I’m going to teach you guys how to teach
computers so that’s not exactly a
typical hobby that most of us go into
it’s not like you finished class and you
just go home and decide to start
teaching computers it’s not exactly
something we all do but for me it all
started when I went on a trip to
Afghanistan as a 15 year old and I came
back and I realized that I wanted to
revolutionize healthcare now that’s a
daunting task it’s a daunting task for a
lot of people and there’s not exactly a
how-to or a wiki how that’s they going
about it there’s no video that explains
step by step instructions nothing
especially for 15 year old me I knew
that that’s what I wanted to do but I
had no clue how I was gonna go about it
until one day driven by my curiosity I
searched something how did that
supercomputer beat the Jeopardy world
champion now in 2011 IBM released Watson
a computer that had question-answering
capabilities now IBM claimed that Watson
was so good at answering questions that
it could beat the world champions in
jeopardy now Jeopardy is a pretty
difficult game it’s not exactly
something that people that aren’t
exactly intelligent player do well in
but for IBM to have the gusto to say
that computers could beat humans
that’s pretty atypical of them and so
when the time came for Watson to go up
against these humans
Watson completely destroyed them there
was not even a close shot for the humans
not only did Watson beat one human it
beat two of the world’s best jeopardy
players and it beat him by a longshot
and so as I’m going through the results
I’m sort of wondering how are computers
better than humans when did that
when did this happen what exactly makes
this computer better than a human at
answering a question and so I’m going
through these Ling’s I’m searching this
weird concept of machine learning comes
up every link has something to do with
machine learning and so I asked how does
machine learning work what is it and so
the basic concept of machine learning is
sort of in the named machines that can
learn it’s a large sort of branch in
computer science and what it entails is
that we can feed computers a ton of data
and we can let these computers figure
out high dimensional patterns clustering
capabilities within this data and if you
teach it well enough these computers get
sort of intelligent at this task and
they can actually take what they’ve
learned and apply it to new problems
that it has never seen before so machine
learning is this topic within computer
science and there are sub branches under
it there are different types of
algorithms that machine learning uses
one of the algorithms that were being
used were called neural networks and
they peaked my interest mainly because
they were kicking human but they were
better than humans at almost every task
that they were being put at and they
were consistently outperforming every
other algorithm that was insane and so
neural networks are modeled sort of like
the human brain that’s where the term
neural comes in we have neurons in our
brain and each of these neurons do
something they have some sort of
function so that’s sort of what in your
own networks do we give it a ton of data
and it sort of sort of figures out what
exactly needs to be looking for but I
wondered what specifically our neural
networks see good at and what I realize
going through these searches and going
through these links was that neural
networks were really good at taking
large amounts of data just straight
number data and finding these obscure
patterns that as humans we would never
realize that were really good at looking
at patterns at such a different
standpoint at high dimensions that as
humans we just can’t comprehend and so
as I’m looking through this and I’m
thinking about this I’m
wondering well regular data regular
numeric data is cool right it works I
mean obviously computers can handle it
but what about images so I asked do
neural networks work with images now
here’s the thing with images you guys
might not realize it or maybe you do but
if you take a picture of me right now
and then you take another picture of me
a second later and then 100 more
pictures of me a hundred seconds later
it’s gonna be different every single
time the lighting is different my facial
expression is different and probably to
the left or to the right it’s different
images have no sort of how-tos of taking
the same exact image every single time
and so if we really want computers to
get good at images they really need to
be learning specifically the content
within the image not environmental
heuristics and so as I’m going through
this neural networks are also a branch
of machine learning but they also
encompass a lot more there are specific
types of neural networks neural networks
that work well with images are called
convolutional neural networks and
essentially instead of having small
mathematical nodes they have these
filters so I sort of understood how
convolutional neural networks worked I
understood that they were good with
images I understood that the concept of
neural networks were good with
extracting high dimensional pattern data
so I wondered how exactly do you train
it like what goes into that and so I
asked how do you train a convolution
your own network and this was astounding
to see because it’s so similar as to how
we as humans are learning you give a
convolutional neural network a ton of
data and this data is labeled so a given
Network a picture of an orange a picture
of an apple another picture of an orange
another picture of an apple and I’ll do
this a couple million times and I’ll
tell the network here’s an image and
guess what you think this image might be
and it’ll take a guess yeah apple or
orange and if it’s wrong I tell it hey
go back and sort of figure out where you
made your mistake tweak it so that if I
were to give you the same exact image
you wouldn’t get it wrong and if it was
right it would hardwire that the
knowledge that it’s been taught sort of
like the brain trial-and-error you get a
math problem wrong you go back you
figure out what you did wrong with it
and then you continue but I I wondered
what exactly within the network are you
training the stuff gets super complex
like is it just is it literally just a
network is that a bunch of code like
what exactly are you training with the
within a convolutional neural network
and I asked what is a convolutional
filter that’s the stuff inside of it and
so these filters are sort of like
Instagram filters Instagram filters make
us look happier and make us look more
energetic sometimes they make me look
like I took my picture and good lighting
when actuality I took it in the bathroom
and dark but these filters aren’t making
us look good they’re looking for
specific features they’re looking for
edges are looking for corners filters
that are later on in the network are
looking at colors and specific objects
if you were to peer within a
convolutional neural network that was
trained to recognize cats and dogs
the last convolutional filter is
actually looking at a cat it’s looking
for a whole facial description of a cat
which is intense because we didn’t
explicitly tell it this is what a cat is
we just let it figure it out after
hundreds of thousands of images and so I
also wondered because this stuff relates
so much to the brain our convolutional
neural networks sort of like the visual
cortex in our brain and it was
interesting to see that they’re almost
exactly the same instead of these
filters we have layers of neurons in our
brain and these neurons are picking up
on the most minut details if I were to
show you a 90 degree angle on this
screen there are specific neurons in
your brain that are going to fire
because they just recognize that 90
degree angle but then if I show off an
89 degree angle those neurons aren’t
going to fire and it set another set of
neurons are going to fire because our
brains are so adept at taking these
really small minut details and figuring
it out
understand
that this is a corner and this is an
edge and this is a 90-degree angle and
this isn’t so as I’m going through this
I sort of have this idea these networks
are really good with patterns with data
and so my next question was about breast
cancer
I wondered why is breast cancer so hard
to diagnose and when I later found out
was that there are a ton of markers for
breast cancer for a human for a
radiologist a pathologist or a doctor to
look at an image an x-ray there are over
10 to 20 features that they need to be
looking for but they need to look for it
within the whole image and they’re not
exactly clustered together these are
things like shape and size and
jaggedness of whatever they’re looking
at and so it’s pretty hard to diagnose
because the difference between a tumor
that’s cancerous and could actually be
fatal
versus a tumor that’s non cancerous and
is really just gonna sit there there’s
almost that much of change there’s not
that much for them to be looking for so
I wanted to quantify that it’s hard to
diagnose but how hard is it and so I
asked what is the misdiagnosis rate so
I’m scrolling through this and I’m
looking at stats and it looks like as
many as 31 percent of women are being
misdiagnosed 31 percent that means that
31 percent of women are being told that
the tumors they have are cancerous when
they’re not actually we’re more fatally
they’re being told that the tumors they
have are not cancerous when they
actually are 31 percent of women are
fighting for their lives they’re
grappling with this reality that there’s
a foreign object within their body that
their body has turned against themselves
and 31 percent of them are being told
they’re wrong that it’s off that’s 31
percent of woman that could be saved
because especially with cancer the
earlier you detect it the higher the
likelihood you have of beating it so
we’re robbing 31% of women
from the chance of living so I’m
thinking about this and it’s astounding
because as I’m going through this the 31
percent is a big number but bigger than
that is the astounding number of threads
that I’m looking at that talked about
how they were misdiagnosed and now
there’s stage for an invasive those are
big words you don’t have to be a doctor
to realize that that doesn’t sound good
and so I wanted to build an accurate
solution something that could help these
people but I didn’t want it to be
expensive so I asked are mammograms
cheap I wanted to build a solution that
was cost effective for people all around
the world and what I realized was that
mammograms are usually covered under
health insurance but if they’re not most
developing countries have these machines
because they’ve been in existence for so
long and so that’s exactly what I did I
took what I learned in computer science
and machine learning in pattern
recognition and I applied it to medicine
to help try to make impact to go towards
that goal of revolutionising healthcare
but what I realized along the way was
that there are just some things that you
can’t search for the passion the
motivation the human element the thing
that wakes us up that keeps us going
these are things that no company no
curriculum no teacher not even your
parents can tell you these are things
that you have to go out into the real
world and explore and find for me I was
going to Afghanistan and realizing that
there were people suffering because the
quality of health care they were
receiving was dictated by the amount of
money that they had health care is the
industry that keeps us living it keeps
us going we have dreams we have
motivations we want to get things done
we have goals we want to we want to grow
up
but when healthcare the industry that is
supposed to keep us living is dictated
by money by wealth by these material
things are our lives really immaterial
is it fair to tell someone that was born
into a situation so
poverty in a thorough country like
Afghanistan where I’m from but because
you were born into poverty your life
doesn’t matter as much compared to the
person that has more money has more
wealth has more influence is that fair
that was a question I grappled with I
realized that this was against every
bone in my body but this experience I
went out of my comfort zone I left the
privilege of living in Chicago the
privilege of being able to go to the
doctor and understand that I was going
to be okay and I saw that there were
woman mothers cradling their sons in
hopes in fear that a doctor would come
by that the line that stood in front of
them the one that grasps corners and
blocks of the city that it would be
shortened so if their little baby
could have a chance to grow up to keep
going it stood against everything I
believed in instead against it and it
shook me and so I came back and I
realized that that was something that I
wanted to fix these things you can’t
find them in a classroom you can’t find
them sitting behind a computer searching
for something you have to go out you
have to explore you have to find the
thing that matters to you I’m blessed to
be able to wake up every morning and
know that I’m happy that I’m content
with the work that I’m doing life is
worth a lot more than money it’s worth a
lot more than what we placed on it in
these developing countries people can’t
control what they’re born into and they
can’t really control where even how they
might end but they can control
everything that happens in the middle
and sometimes we get lucky we’re a lot
luckier because we’re born into a place
where we can do more and it’s our
responsibility as humans forget forget
that it’s them these are our brothers or
sisters these are people that we see on
the news and we just assume oh it’s
someone else
that country but no these are real
people that have families that have
goals and motivations just like us and
so I want to ask you the audience the
people sitting here you guys want to
make bold moves that’s why you’re here
you’re in support of that ideology what
exactly is it that you want to search
for go out do something find something
worth searching for thank you
[Applause]
you
[Applause]
Please follow and like us: