the field of artificial intelligence has been moving extremely quickly in the
been moving extremely quickly in the
been moving extremely quickly in the last few years do you know how AI will
last few years do you know how AI will
last few years do you know how AI will affect your life in the future I have
affect your life in the future I have
affect your life in the future I have some ideas
some ideas
some ideas I’m Jeff Dean I’m a Google senior fellow
I’m Jeff Dean I’m a Google senior fellow
I’m Jeff Dean I’m a Google senior fellow at Google that means a couple of things
at Google that means a couple of things
at Google that means a couple of things first means I’m kind of old second it
first means I’m kind of old second it
first means I’m kind of old second it means that I get to spend my time
means that I get to spend my time
means that I get to spend my time working on problems that I think are the
working on problems that I think are the
working on problems that I think are the most important for the company I’m
most important for the company I’m
most important for the company I’m working on artificial intelligence I
working on artificial intelligence I
working on artificial intelligence I lead the Google brain team our
lead the Google brain team our
lead the Google brain team our artificial intelligence research team
artificial intelligence research team
artificial intelligence research team based in Mountain View California our
based in Mountain View California our
based in Mountain View California our team does long-term research on how to
team does long-term research on how to
team does long-term research on how to make machines intelligent and we then
make machines intelligent and we then
make machines intelligent and we then work with product teams at Google in
work with product teams at Google in
work with product teams at Google in order to make machine products that are
order to make machine products that are
order to make machine products that are intelligent we want to use intelligence
intelligent we want to use intelligence
intelligent we want to use intelligence to augment the abilities of people to
to augment the abilities of people to
to augment the abilities of people to enable us to accomplish more to
enable us to accomplish more to
enable us to accomplish more to eliminate tedious repetitive tasks and
eliminate tedious repetitive tasks and
eliminate tedious repetitive tasks and to and allow us to spend more time on
to and allow us to spend more time on
to and allow us to spend more time on our creative endeavors ai is going to be
our creative endeavors ai is going to be
our creative endeavors ai is going to be more impactful than the invention of the
more impactful than the invention of the
more impactful than the invention of the personal computer and the spread of
personal computer and the spread of
personal computer and the spread of mobile phones into your pocket so the
mobile phones into your pocket so the
mobile phones into your pocket so the idea of artificial intelligence is not
idea of artificial intelligence is not
idea of artificial intelligence is not new it’s been around since the very
new it’s been around since the very
new it’s been around since the very earliest days of computing it’s a grand
earliest days of computing it’s a grand
earliest days of computing it’s a grand project to build machines that are
project to build machines that are
project to build machines that are intelligent and there’s many ways of
intelligent and there’s many ways of
intelligent and there’s many ways of pursuing this its captivated computer
pursuing this its captivated computer
pursuing this its captivated computer scientists ever since the most promising
scientists ever since the most promising
scientists ever since the most promising approach though is the area of machine
approach though is the area of machine
approach though is the area of machine learning rather than trying to embody
learning rather than trying to embody
learning rather than trying to embody machines with everything they need to
machines with everything they need to
machines with everything they need to know up front rather we want to enable
know up front rather we want to enable
know up front rather we want to enable them to learn to learn how to learn so
them to learn to learn how to learn so
them to learn to learn how to learn so that they can learn from their
that they can learn from their
that they can learn from their observations of the world and to make
observations of the world and to make
observations of the world and to make inferences based on those observations
inferences based on those observations
inferences based on those observations the field of deep learning is a
the field of deep learning is a
the field of deep learning is a particular kind of machine learning that
particular kind of machine learning that
particular kind of machine learning that I’ll be talking about and it’s been
I’ll be talking about and it’s been
I’ll be talking about and it’s been shown in the last four or five years to
shown in the last four or five years to
shown in the last four or five years to be remarkably effective for a wide
be remarkably effective for a wide
be remarkably effective for a wide variety of problems although it’s
variety of problems although it’s
variety of problems although it’s actually much older before we start
actually much older before we start
actually much older before we start though how do we learn why we learn from
though how do we learn why we learn from
though how do we learn why we learn from examples of things and from repeated
examples of things and from repeated
examples of things and from repeated practice and repeated practice and
practice and repeated practice and
practice and repeated practice and Impuls are really key to machine
Impuls are really key to machine
Impuls are really key to machine learning as well so in machine learning
learning as well so in machine learning
learning as well so in machine learning we’re going to expose a system to
we’re going to expose a system to
we’re going to expose a system to examples of the behavior we want it to
examples of the behavior we want it to
examples of the behavior we want it to have and those examples are going to
have and those examples are going to
have and those examples are going to teach it it’s going to learn from those
teach it it’s going to learn from those
teach it it’s going to learn from those examples how to do something so in this
examples how to do something so in this
examples how to do something so in this very simple diagram I have a model where
very simple diagram I have a model where
very simple diagram I have a model where we’re going to try to teach a computer
we’re going to try to teach a computer
we’re going to try to teach a computer to tell whether a photograph contains a
to tell whether a photograph contains a
to tell whether a photograph contains a cat or a dog and we’re going to have
cat or a dog and we’re going to have
cat or a dog and we’re going to have examples where we know the right answer
examples where we know the right answer
examples where we know the right answer so we have some that we say that’s a cat
so we have some that we say that’s a cat
so we have some that we say that’s a cat that’s a dog that’s a dog that’s a cat
that’s a dog that’s a dog that’s a cat
that’s a dog that’s a dog that’s a cat and we’re going to essentially show the
and we’re going to essentially show the
and we’re going to essentially show the computer an example and say what do you
computer an example and say what do you
computer an example and say what do you think they’re going to say well I think
think they’re going to say well I think
think they’re going to say well I think that’s a dog and there you go if you get
that’s a dog and there you go if you get
that’s a dog and there you go if you get the right answer you’re done right just
the right answer you’re done right just
the right answer you’re done right just like doing your homework problems if you
like doing your homework problems if you
like doing your homework problems if you get it right you don’t have to do more
get it right you don’t have to do more
get it right you don’t have to do more but if you get it wrong we’re going to
but if you get it wrong we’re going to
but if you get it wrong we’re going to make little adjustments to how we
make little adjustments to how we
make little adjustments to how we believe about what we believe about the
believe about what we believe about the
believe about what we believe about the world and what we believe about this
world and what we believe about this
world and what we believe about this example to make it more likely that the
example to make it more likely that the
example to make it more likely that the next time we see that example or one
next time we see that example or one
next time we see that example or one like it will get the right answer not
like it will get the right answer not
like it will get the right answer not the right wrong one deep learning in
the right wrong one deep learning in
the right wrong one deep learning in particular has a particular way of doing
particular has a particular way of doing
particular has a particular way of doing this that is really important so it
this that is really important so it
this that is really important so it builds these layers of abstraction
builds these layers of abstraction
builds these layers of abstraction automatically as part of the learning
automatically as part of the learning
automatically as part of the learning process where the lowest level things or
process where the lowest level things or
process where the lowest level things or things like you know does this part of
things like you know does this part of
things like you know does this part of the image contain a little splotch of
the image contain a little splotch of
the image contain a little splotch of brown and then as you go up through the
brown and then as you go up through the
brown and then as you go up through the layers blub why are the kinds of things
layers blub why are the kinds of things
layers blub why are the kinds of things that it learns get more complicated
that it learns get more complicated
that it learns get more complicated things like is there an ear at this part
things like is there an ear at this part
things like is there an ear at this part of the picture or something looks like a
of the picture or something looks like a
of the picture or something looks like a couple of eyes or maybe some whiskers
couple of eyes or maybe some whiskers
couple of eyes or maybe some whiskers and those features emerge automatically
and those features emerge automatically
and those features emerge automatically as part of the learning process which is
as part of the learning process which is
as part of the learning process which is really critical aspect of deep learning
really critical aspect of deep learning
really critical aspect of deep learning we don’t have to tell it how to tell a
we don’t have to tell it how to tell a
we don’t have to tell it how to tell a cat from a dog it just learns that
cat from a dog it just learns that
cat from a dog it just learns that there’s these things called whiskers and
there’s these things called whiskers and
there’s these things called whiskers and they seem to appear in lots of photos
they seem to appear in lots of photos
they seem to appear in lots of photos they seem to appear more often and ones
they seem to appear more often and ones
they seem to appear more often and ones about cats and that’s really helpful and
about cats and that’s really helpful and
about cats and that’s really helpful and those features develop automatically in
those features develop automatically in
those features develop automatically in the learning process so neural nets
the learning process so neural nets
the learning process so neural nets cannot can learn much more than just how
cannot can learn much more than just how
cannot can learn much more than just how do you tell a cat from a dog they can
do you tell a cat from a dog they can
do you tell a cat from a dog they can learn to distinguish thousands of
learn to distinguish thousands of
learn to distinguish thousands of different categories of objects you know
different categories of objects you know
different categories of objects you know tens of thousands this is an ostrich or
tens of thousands this is an ostrich or
tens of thousands this is an ostrich or a fire truck or a fire boat things like
a fire truck or a fire boat things like
a fire truck or a fire boat things like that they can learn from an audio stream
that they can learn from an audio stream
that they can learn from an audio stream to give you a transcript of the words
to give you a transcript of the words
to give you a transcript of the words that were said in that audio stream how
that were said in that audio stream how
that were said in that audio stream how cold is it outside they can take in an
cold is it outside they can take in an
cold is it outside they can take in an English sentence hello
English sentence hello
English sentence hello how are you and spit out the
how are you and spit out the
how are you and spit out the corresponding French sentence Bonjour
corresponding French sentence Bonjour
corresponding French sentence Bonjour comment allez-vous sorry for my French
they can take in the pixels of an image
they can take in the pixels of an image and give you more than just a category
and give you more than just a category
and give you more than just a category about it they can actually write a
about it they can actually write a
about it they can actually write a sentence a caption if you will about
sentence a caption if you will about
sentence a caption if you will about that picture a blue and yellow train
that picture a blue and yellow train
that picture a blue and yellow train travelling down the train tracks that
travelling down the train tracks that
travelling down the train tracks that perch hose a pretty high level of
perch hose a pretty high level of
perch hose a pretty high level of understanding of what’s going on in that
understanding of what’s going on in that
understanding of what’s going on in that scene one of the great things about deep
scene one of the great things about deep
scene one of the great things about deep learning is that all of those things
learning is that all of those things
learning is that all of those things that I showed you can be expressed using
that I showed you can be expressed using
that I showed you can be expressed using a relatively simple set of algorithms
a relatively simple set of algorithms
a relatively simple set of algorithms and can be expressed in a common
and can be expressed in a common
and can be expressed in a common software framework so we can build a
software framework so we can build a
software framework so we can build a software framework that enables us to
software framework that enables us to
software framework that enables us to express all these different learning
express all these different learning
express all these different learning problems and then use it over and over
problems and then use it over and over
problems and then use it over and over for our research and for our products
for our research and for our products
for our research and for our products and so the system we built this called
and so the system we built this called
and so the system we built this called tensorflow and we use it internally for
tensorflow and we use it internally for
tensorflow and we use it internally for everything that we do for in these this
everything that we do for in these this
everything that we do for in these this area and last year we decided we would
area and last year we decided we would
area and last year we decided we would open-source it because we wanted people
open-source it because we wanted people
open-source it because we wanted people to have the ability to take this
to have the ability to take this
to have the ability to take this software download it for free and use it
software download it for free and use it
software download it for free and use it for their learning problems it’s been
for their learning problems it’s been
for their learning problems it’s been really great to see different things
really great to see different things
really great to see different things that people have used it for so here’s
that people have used it for so here’s
that people have used it for so here’s one example there was a Japanese
one example there was a Japanese
one example there was a Japanese cucumber farmer and it turns out in when
cucumber farmer and it turns out in when
cucumber farmer and it turns out in when you harvest your cucumbers you have to
you harvest your cucumbers you have to
you harvest your cucumbers you have to sort them into all kinds of different
sort them into all kinds of different
sort them into all kinds of different categories for sale small ones medium
categories for sale small ones medium
categories for sale small ones medium ones large ones
ones large ones
ones large ones prickly ones NOT prickly ones straight
prickly ones NOT prickly ones straight
prickly ones NOT prickly ones straight ones curved ones this is pretty
ones curved ones this is pretty
ones curved ones this is pretty complicated I’m pretty time consuming at
complicated I’m pretty time consuming at
complicated I’m pretty time consuming at harvest time so the farmer was able to
harvest time so the farmer was able to
harvest time so the farmer was able to take a camera and using a computer
take a camera and using a computer
take a camera and using a computer vision model that he trained with
vision model that he trained with
vision model that he trained with tensorflow actually have the vision
tensorflow actually have the vision
tensorflow actually have the vision model determine what category of
model determine what category of
model determine what category of cucumber I was looking at and then
cucumber I was looking at and then
cucumber I was looking at and then rigged it up to some conveyor belts and
rigged it up to some conveyor belts and
rigged it up to some conveyor belts and some little switches that would push the
some little switches that would push the
some little switches that would push the cucumber into the right box and so this
cucumber into the right box and so this
cucumber into the right box and so this eliminated many days of labor that the
eliminated many days of labor that the
eliminated many days of labor that the farmer and his wife would have to do at
farmer and his wife would have to do at
farmer and his wife would have to do at harvest time that’s just one tiny
harvest time that’s just one tiny
harvest time that’s just one tiny example of something you can do now
example of something you can do now
example of something you can do now that would be harder before so as I said
that would be harder before so as I said
that would be harder before so as I said neural nets are not new they’ve actually
neural nets are not new they’ve actually
neural nets are not new they’ve actually been around since the 1980s and 1990s
been around since the 1980s and 1990s
been around since the 1980s and 1990s and they showed really promising results
and they showed really promising results
and they showed really promising results on kind of small toy problems then but
on kind of small toy problems then but
on kind of small toy problems then but they really couldn’t show great great
they really couldn’t show great great
they really couldn’t show great great results on realistic large problems at
results on realistic large problems at
results on realistic large problems at that time and the reason is we just
that time and the reason is we just
that time and the reason is we just lacked enough computational power the
lacked enough computational power the
lacked enough computational power the process of making these adjustments to
process of making these adjustments to
process of making these adjustments to the model on every example and
the model on every example and
the model on every example and processing every example many times in
processing every example many times in
processing every example many times in order to build this model of the world
order to build this model of the world
order to build this model of the world is very computationally intensive and so
is very computationally intensive and so
is very computationally intensive and so we just needed faster computers
we just needed faster computers
we just needed faster computers fortunately we have faster computers now
fortunately we have faster computers now
fortunately we have faster computers now computers have been dramatically
computers have been dramatically
computers have been dramatically improving in performance every year for
improving in performance every year for
improving in performance every year for the last 30 to 40 years and we’ve
the last 30 to 40 years and we’ve
the last 30 to 40 years and we’ve reached this point where neural nets are
reached this point where neural nets are
reached this point where neural nets are actually practical on real problems so
actually practical on real problems so
actually practical on real problems so the computer you have in your phone is
the computer you have in your phone is
the computer you have in your phone is now a hundred to a thousand times as
now a hundred to a thousand times as
now a hundred to a thousand times as powerful as the computers you had on
powerful as the computers you had on
powerful as the computers you had on your desktop twenty or thirty years ago
your desktop twenty or thirty years ago
your desktop twenty or thirty years ago and that makes all the difference so we
and that makes all the difference so we
and that makes all the difference so we now have enough compute computation and
now have enough compute computation and
now have enough compute computation and just to take an example the field with
just to take an example the field with
just to take an example the field with computer vision every year there’s a
computer vision every year there’s a
computer vision every year there’s a contest where teams compete to see who
contest where teams compete to see who
contest where teams compete to see who can give the right categories out of a
can give the right categories out of a
can give the right categories out of a thousand different categories when given
thousand different categories when given
thousand different categories when given an image and in 2011 before people were
an image and in 2011 before people were
an image and in 2011 before people were using neural nets
using neural nets
using neural nets the winning team got an error rate of
the winning team got an error rate of
the winning team got an error rate of 26% which doesn’t sound too good when
26% which doesn’t sound too good when
26% which doesn’t sound too good when you think that humans are at 5% on this
you think that humans are at 5% on this
you think that humans are at 5% on this task but fast-forward five just five
task but fast-forward five just five
task but fast-forward five just five years and we’re now at three percent
years and we’re now at three percent
years and we’re now at three percent errors using deep learning and much more
errors using deep learning and much more
errors using deep learning and much more computational power
computational power
computational power we’re actually better than humans on
we’re actually better than humans on
we’re actually better than humans on this task that’s a really powerful and
this task that’s a really powerful and
this task that’s a really powerful and transformative thing think about this
transformative thing think about this
transformative thing think about this computers can now see and they didn’t
computers can now see and they didn’t
computers can now see and they didn’t used to be able to if you think back to
used to be able to if you think back to
used to be able to if you think back to the time and evolutionary biology when
the time and evolutionary biology when
the time and evolutionary biology when computer when animals of all dies that
computer when animals of all dies that
computer when animals of all dies that was likely a time of great change and
was likely a time of great change and
was likely a time of great change and incredible amazing things started to
incredible amazing things started to
incredible amazing things started to happen and computer we’re now at that
happen and computer we’re now at that
happen and computer we’re now at that point in computing for example this is
point in computing for example this is
point in computing for example this is really really useful if you’re trying to
really really useful if you’re trying to
really really useful if you’re trying to build a robot if you can’t see is really
build a robot if you can’t see is really
build a robot if you can’t see is really hard to do stuff
hard to do stuff
hard to do stuff so here we have an example of robots
so here we have an example of robots
so here we have an example of robots using deep learning to teach themselves
using deep learning to teach themselves
using deep learning to teach themselves hand-eye coordination so essentially
hand-eye coordination so essentially
hand-eye coordination so essentially there’s a video camera for each robot
there’s a video camera for each robot
there’s a video camera for each robot gets to look over its shoulder I guess I
gets to look over its shoulder I guess I
gets to look over its shoulder I guess I don’t know it only has one arm but over
don’t know it only has one arm but over
don’t know it only has one arm but over its shoulder and again that model is
its shoulder and again that model is
its shoulder and again that model is going to take the pixel inputs from the
going to take the pixel inputs from the
going to take the pixel inputs from the camera and go directly to six torque
camera and go directly to six torque
camera and go directly to six torque motor commands for the different joints
motor commands for the different joints
motor commands for the different joints of the robot and they’re essentially
of the robot and they’re essentially
of the robot and they’re essentially just going to through trial and error
just going to through trial and error
just going to through trial and error practice picking things up and so you
practice picking things up and so you
practice picking things up and so you see they’re just trying things and they
see they’re just trying things and they
see they’re just trying things and they can tell if they succeeded or failed by
can tell if they succeeded or failed by
can tell if they succeeded or failed by whether their gripper closed all the way
whether their gripper closed all the way
whether their gripper closed all the way or whether they were actually successful
or whether they were actually successful
or whether they were actually successful at picking this up and they’re learning
at picking this up and they’re learning
at picking this up and they’re learning like what kinds of grips work well for
like what kinds of grips work well for
like what kinds of grips work well for different shape objects based on the
different shape objects based on the
different shape objects based on the vision that they’re there were
vision that they’re there were
vision that they’re there were developing and they’re pretty good at it
developing and they’re pretty good at it
developing and they’re pretty good at it we bought a bunch of variety packs of
we bought a bunch of variety packs of
we bought a bunch of variety packs of toys and tools on Amazon another area
toys and tools on Amazon another area
toys and tools on Amazon another area where I think machine learning has
where I think machine learning has
where I think machine learning has incredible opportunities is in the area
incredible opportunities is in the area
incredible opportunities is in the area of healthcare and I’ll give you just one
of healthcare and I’ll give you just one
of healthcare and I’ll give you just one example
so diabetic retinopathy is the fastest
so diabetic retinopathy is the fastest growing cause of blindness in the world
growing cause of blindness in the world
growing cause of blindness in the world there’s 400 million people at risk who
there’s 400 million people at risk who
there’s 400 million people at risk who should really get screened basically
should really get screened basically
should really get screened basically every year but but often people don’t
every year but but often people don’t
every year but but often people don’t get the screening they should so we
get the screening they should so we
get the screening they should so we wanted to tackle this with computer
wanted to tackle this with computer
wanted to tackle this with computer vision and you get an image like this
vision and you get an image like this
vision and you get an image like this and traditionally an ophthalmologist
and traditionally an ophthalmologist
and traditionally an ophthalmologist scans this to try to assess how serious
scans this to try to assess how serious
scans this to try to assess how serious the signs are if there are if there are
the signs are if there are if there are
the signs are if there are if there are any and so we got a large collection of
any and so we got a large collection of
any and so we got a large collection of these eye images and we had human
these eye images and we had human
these eye images and we had human ophthalmologist labeled them so if you
ophthalmologist labeled them so if you
ophthalmologist labeled them so if you have a human ophthalmologist to human
have a human ophthalmologist to human
have a human ophthalmologist to human ophthalmologists score these they agree
ophthalmologists score these they agree
ophthalmologists score these they agree with their rating sixty percent of the
with their rating sixty percent of the
with their rating sixty percent of the time
time
time slightly more worrisome though is if you
slightly more worrisome though is if you
slightly more worrisome though is if you ask the same ophthalmologist to grade
ask the same ophthalmologist to grade
ask the same ophthalmologist to grade the same image a few hours later they
the same image a few hours later they
the same image a few hours later they only agree with themselves 65 percent of
only agree with themselves 65 percent of
only agree with themselves 65 percent of the time and it’s really just a hard
the time and it’s really just a hard
the time and it’s really just a hard problem right it’s sort of an
problem right it’s sort of an
problem right it’s sort of an interpretation of like how how dark are
interpretation of like how how dark are
interpretation of like how how dark are those spots should that be a 2 or a 3 in
those spots should that be a 2 or a 3 in
those spots should that be a 2 or a 3 in this rating scale so on so in work
this rating scale so on so in work
this rating scale so on so in work published earlier this week by our group
published earlier this week by our group
published earlier this week by our group in the Journal of the American Medical
in the Journal of the American Medical
in the Journal of the American Medical station we now have a machine learning
station we now have a machine learning
station we now have a machine learning model that performs on par perhaps even
model that performs on par perhaps even
model that performs on par perhaps even slightly better than ophthalmologists at
slightly better than ophthalmologists at
slightly better than ophthalmologists at this task and this is really important
this task and this is really important
this task and this is really important because it can make ophthalmologists
because it can make ophthalmologists
because it can make ophthalmologists much more efficient they can actually
much more efficient they can actually
much more efficient they can actually dedicate their time to the people who
dedicate their time to the people who
dedicate their time to the people who deserve the attention and not spend as
deserve the attention and not spend as
deserve the attention and not spend as much time on screening screening most of
much time on screening screening most of
much time on screening screening most of the people most of whom don’t don’t have
the people most of whom don’t don’t have
the people most of whom don’t don’t have any signs of this will also be more
any signs of this will also be more
any signs of this will also be more creative by having tools that understand
creative by having tools that understand
creative by having tools that understand the art we’re trying to create that will
the art we’re trying to create that will
the art we’re trying to create that will be really helpful I’ll show you another
be really helpful I’ll show you another
be really helpful I’ll show you another example so Leon Gatti’s and his
example so Leon Gatti’s and his
example so Leon Gatti’s and his colleagues from the University of
colleagues from the University of
colleagues from the University of Tuebingen
Tuebingen
Tuebingen and the Max Planck Institute in Germany
and the Max Planck Institute in Germany
and the Max Planck Institute in Germany last year published this amazing paper
last year published this amazing paper
last year published this amazing paper where they have an algorithm that can
where they have an algorithm that can
where they have an algorithm that can take two things an image photograph and
take two things an image photograph and
take two things an image photograph and a painting and what the algorithm does
a painting and what the algorithm does
a painting and what the algorithm does is it renders that photograph in the
is it renders that photograph in the
is it renders that photograph in the style of that painter so here you see
style of that painter so here you see
style of that painter so here you see that same picture rendered in three
that same picture rendered in three
that same picture rendered in three different styles automatically by those
different styles automatically by those
different styles automatically by those three different artists that’s pretty
three different artists that’s pretty
three different artists that’s pretty amazing and I think there’s going to be
amazing and I think there’s going to be
amazing and I think there’s going to be a real opportunity for to create tools
a real opportunity for to create tools
a real opportunity for to create tools for human artists that allow them to
for human artists that allow them to
for human artists that allow them to really interact with systems like this
really interact with systems like this
really interact with systems like this and to more rapidly get the ideas in
and to more rapidly get the ideas in
and to more rapidly get the ideas in their head out into really new and
their head out into really new and
their head out into really new and creative kinds of art remember how the
creative kinds of art remember how the
creative kinds of art remember how the advance from the 1980s to now was caused
advance from the 1980s to now was caused
advance from the 1980s to now was caused by much more computational power so the
by much more computational power so the
by much more computational power so the same thing is going to be true in the
same thing is going to be true in the
same thing is going to be true in the future we want more computational power
future we want more computational power
future we want more computational power so that we can train larger models so we
so that we can train larger models so we
so that we can train larger models so we can learn more and so deep learning is
can learn more and so deep learning is
can learn more and so deep learning is actually transforming how we design and
actually transforming how we design and
actually transforming how we design and build computers as well there’s two
build computers as well there’s two
build computers as well there’s two interesting properties that neural nets
interesting properties that neural nets
interesting properties that neural nets and learning algorithms were using have
and learning algorithms were using have
and learning algorithms were using have the first is that reduced precision is
the first is that reduced precision is
the first is that reduced precision is okay it’s fine when you’re multiplying
okay it’s fine when you’re multiplying
okay it’s fine when you’re multiplying numbers for a null net to say about 1.2
numbers for a null net to say about 1.2
numbers for a null net to say about 1.2 times about point 6 hmm sounds about
times about point 6 hmm sounds about
times about point 6 hmm sounds about like point 7 that’s good enough we don’t
like point 7 that’s good enough we don’t
like point 7 that’s good enough we don’t have to spend every last detail of our
have to spend every last detail of our
have to spend every last detail of our computational budget preserving all the
computational budget preserving all the
computational budget preserving all the digits of precision that traditional
digits of precision that traditional
digits of precision that traditional CPUs and
CPUs and
CPUs and are designed to build and that that’s
are designed to build and that that’s
are designed to build and that that’s really helpful if you just you yourself
really helpful if you just you yourself
really helpful if you just you yourself if you needed to multiply a bunch of
if you needed to multiply a bunch of
if you needed to multiply a bunch of things and you were able to be very
things and you were able to be very
things and you were able to be very approximate like that you could do many
approximate like that you could do many
approximate like that you could do many more operations same thing is happening
more operations same thing is happening
more operations same thing is happening in neural nets the other thing is that
in neural nets the other thing is that
in neural nets the other thing is that all the learning systems I showed you
all the learning systems I showed you
all the learning systems I showed you and all the algorithms rely on just a
and all the algorithms rely on just a
and all the algorithms rely on just a handful of specific operations they
handful of specific operations they
handful of specific operations they don’t need the full generality the
don’t need the full generality the
don’t need the full generality the general-purpose computers have they want
general-purpose computers have they want
general-purpose computers have they want to be able to do things like matrix
to be able to do things like matrix
to be able to do things like matrix multiplies vector operations things from
multiplies vector operations things from
multiplies vector operations things from linear algebra and that’s about it and
linear algebra and that’s about it and
linear algebra and that’s about it and so that allows us to build specialized
so that allows us to build specialized
so that allows us to build specialized computers that can do these things
computers that can do these things
computers that can do these things extremely well and not much else so here
extremely well and not much else so here
extremely well and not much else so here is a system that we’ve built at Google
is a system that we’ve built at Google
is a system that we’ve built at Google about in the last three years called the
about in the last three years called the
about in the last three years called the tensor processing unit it’s essentially
tensor processing unit it’s essentially
tensor processing unit it’s essentially a custom-designed chip that accelerates
a custom-designed chip that accelerates
a custom-designed chip that accelerates real night computations only but because
real night computations only but because
real night computations only but because these kinds of deep learning algorithms
these kinds of deep learning algorithms
these kinds of deep learning algorithms are so applicable to many problems this
are so applicable to many problems this
are so applicable to many problems this is great because we now have something
is great because we now have something
is great because we now have something can speed up those kinds of computations
can speed up those kinds of computations
can speed up those kinds of computations by an order of magnitude compared to
by an order of magnitude compared to
by an order of magnitude compared to traditional CPUs and that’s really
traditional CPUs and that’s really
traditional CPUs and that’s really really powerful it allows us to use more
really powerful it allows us to use more
really powerful it allows us to use more powerful models in our products and get
powerful models in our products and get
powerful models in our products and get sort of better systems out there let me
sort of better systems out there let me
sort of better systems out there let me take you on a tour of some queries from
take you on a tour of some queries from
take you on a tour of some queries from the future one of the things that we’ve
the future one of the things that we’ve
the future one of the things that we’ve observed is that as systems get more
observed is that as systems get more
observed is that as systems get more intelligent users expect more from up
intelligent users expect more from up
intelligent users expect more from up from what they can do so which of these
from what they can do so which of these
from what they can do so which of these I Emma jiz show symptoms of diabetic
I Emma jiz show symptoms of diabetic
I Emma jiz show symptoms of diabetic retinopathy well I already showed you we
retinopathy well I already showed you we
retinopathy well I already showed you we can actually do that today
can actually do that today
can actually do that today describe this video in Spanish we can’t
describe this video in Spanish we can’t
describe this video in Spanish we can’t quite do that we can describe still
quite do that we can describe still
quite do that we can describe still images pretty well but not quite moving
images pretty well but not quite moving
images pretty well but not quite moving videos but that’ll come find me
videos but that’ll come find me
videos but that’ll come find me documents related to reinforcement
documents related to reinforcement
documents related to reinforcement learning for robotics and summarize them
learning for robotics and summarize them
learning for robotics and summarize them in German that’s pretty complicated but
in German that’s pretty complicated but
in German that’s pretty complicated but imagine how productive we’ll be able to
imagine how productive we’ll be able to
imagine how productive we’ll be able to be if we’re able to have tools that can
be if we’re able to have tools that can
be if we’re able to have tools that can do that please fetch me a cup of tea
do that please fetch me a cup of tea
do that please fetch me a cup of tea from the kitchen having robots operate
from the kitchen having robots operate
from the kitchen having robots operate in messy environments like your kitchen
in messy environments like your kitchen
in messy environments like your kitchen is actually pretty complicated but the
is actually pretty complicated but the
is actually pretty complicated but the very baby steps of things like the
very baby steps of things like the
very baby steps of things like the hand-eye coordination work I showed you
hand-eye coordination work I showed you
hand-eye coordination work I showed you are step
are step
are step along that path in closing AI is going
along that path in closing AI is going
along that path in closing AI is going to help us to be healthier happier more
to help us to be healthier happier more
to help us to be healthier happier more productive and more creative are you
productive and more creative are you
productive and more creative are you excited to see what the future of AI
excited to see what the future of AI
excited to see what the future of AI holds I really am thank you
holds I really am thank you
holds I really am thank you [Applause]
”
Be First to Comment