good morning this is pretty much about technology and so they gave me this
technology and so they gave me this
technology and so they gave me this piece of technology and I will use it
piece of technology and I will use it
piece of technology and I will use it instead of the clicker and if I pointed
instead of the clicker and if I pointed
instead of the clicker and if I pointed you then please change my slides all
you then please change my slides all
you then please change my slides all right so some things in life are really
right so some things in life are really
right so some things in life are really annoying I guess you all know that and
annoying I guess you all know that and
annoying I guess you all know that and many of these things are connected to
many of these things are connected to
many of these things are connected to these super computers that you carry in
these super computers that you carry in
these super computers that you carry in your back it’s pretty much about
your back it’s pretty much about
your back it’s pretty much about charging the battery this is what you
charging the battery this is what you
charging the battery this is what you have to do every night every evening you
have to do every night every evening you
have to do every night every evening you have to connect this thing and sometimes
have to connect this thing and sometimes
have to connect this thing and sometimes you forget it you want deep trouble the
you forget it you want deep trouble the
you forget it you want deep trouble the next day then there is updating the
next day then there is updating the
next day then there is updating the software there are these constant
software there are these constant
software there are these constant messages reminding us that new fantastic
messages reminding us that new fantastic
messages reminding us that new fantastic features have been implemented bugs have
features have been implemented bugs have
features have been implemented bugs have been removed software became safer and
been removed software became safer and
been removed software became safer and you have to install it and both of these
you have to install it and both of these
you have to install it and both of these things charging your battery and
things charging your battery and
things charging your battery and updating your software are totally
updating your software are totally
updating your software are totally annoying and take away a lot of your
annoying and take away a lot of your
annoying and take away a lot of your productive time every day
productive time every day
productive time every day there is another annoying thing about
there is another annoying thing about
there is another annoying thing about cell phones which many of you or most of
cell phones which many of you or most of
cell phones which many of you or most of you are not aware of it really annoys
you are not aware of it really annoys
you are not aware of it really annoys those who produce these things and these
those who produce these things and these
those who produce these things and these cell phones contain well billions of
cell phones contain well billions of
cell phones contain well billions of transistors and they all have to work
transistors and they all have to work
transistors and they all have to work and these things are really tiny you’re
and these things are really tiny you’re
and these things are really tiny you’re talking about tens of nanometers or ten
talking about tens of nanometers or ten
talking about tens of nanometers or ten nanometers now and imagine you have to
nanometers now and imagine you have to
nanometers now and imagine you have to build a billion or several billion
build a billion or several billion
build a billion or several billion transistors and they all have to work if
transistors and they all have to work if
transistors and they all have to work if one or a few of them fail you can throw
one or a few of them fail you can throw
one or a few of them fail you can throw away the microprocessor which is a very
away the microprocessor which is a very
away the microprocessor which is a very expensive thing and so companies like
expensive thing and so companies like
expensive thing and so companies like Intel TSMC Samsung they have to invest
Intel TSMC Samsung they have to invest
Intel TSMC Samsung they have to invest billions of dollars to produce these
billions of dollars to produce these
billions of dollars to produce these things now if you look our brain the
things now if you look our brain the
things now if you look our brain the three aspects that I just discussed are
three aspects that I just discussed are
three aspects that I just discussed are slightly different rather than charging
slightly different rather than charging
slightly different rather than charging our battery we eat and drink which is a
our battery we eat and drink which is a
our battery we eat and drink which is a lot of fun of course so that’s not a
lot of fun of course so that’s not a
lot of fun of course so that’s not a problem
problem
problem interestingly enough the power
interestingly enough the power
interestingly enough the power that’s the energy per time that we move
that’s the energy per time that we move
that’s the energy per time that we move into our brain is about the same as the
into our brain is about the same as the
into our brain is about the same as the power that we move into our cell phones
power that we move into our cell phones
power that we move into our cell phones it’s a couple of what’s 10 toed about 10
it’s a couple of what’s 10 toed about 10
it’s a couple of what’s 10 toed about 10 watts a bit less for the cell phone a
watts a bit less for the cell phone a
watts a bit less for the cell phone a bit more data for your brain but the
bit more data for your brain but the
bit more data for your brain but the order of magnitude is the same
order of magnitude is the same
order of magnitude is the same another thing is on the right corner
another thing is on the right corner
another thing is on the right corner here is robustness and that is really
here is robustness and that is really
here is robustness and that is really interesting there is no company like
interesting there is no company like
interesting there is no company like Intel or TSMC or Samsung that builds our
Intel or TSMC or Samsung that builds our
Intel or TSMC or Samsung that builds our brain cells but they are growing during
brain cells but they are growing during
brain cells but they are growing during our development and very importantly
our development and very importantly
our development and very importantly they are also disappearing all the time
they are also disappearing all the time
they are also disappearing all the time we all lose about like one brain cell
we all lose about like one brain cell
we all lose about like one brain cell per second that’s a hundred thousand a
per second that’s a hundred thousand a
per second that’s a hundred thousand a day or PI times 10 to the 7 per year
day or PI times 10 to the 7 per year
day or PI times 10 to the 7 per year which is very very substantial of course
which is very very substantial of course
which is very very substantial of course when we we have 10 to the 11 so that’s
when we we have 10 to the 11 so that’s
when we we have 10 to the 11 so that’s not so bad but if you think of a
not so bad but if you think of a
not so bad but if you think of a microprocessor that loses one transistor
microprocessor that loses one transistor
microprocessor that loses one transistor per minute per second ever be a disaster
per minute per second ever be a disaster
per minute per second ever be a disaster it would stop working straightaway so
it would stop working straightaway so
it would stop working straightaway so the brain has a huge amount of whatever
the brain has a huge amount of whatever
the brain has a huge amount of whatever call robustness then most importantly
call robustness then most importantly
call robustness then most importantly there is this thing which are now
there is this thing which are now
there is this thing which are now painted red which is there are no
painted red which is there are no
painted red which is there are no software updates there is no software
software updates there is no software
software updates there is no software that you have to download to your brain
that you have to download to your brain
that you have to download to your brain because it improves your capabilities
because it improves your capabilities
because it improves your capabilities but you do that yourself by a process
but you do that yourself by a process
but you do that yourself by a process called learning sometimes also
called learning sometimes also
called learning sometimes also development and plasticity and you do
development and plasticity and you do
development and plasticity and you do that by interacting with the environment
that by interacting with the environment
that by interacting with the environment so if we could make a system an
so if we could make a system an
so if we could make a system an artificial system that has these
artificial system that has these
artificial system that has these features energy efficiency robustness
features energy efficiency robustness
features energy efficiency robustness and most importantly the ability to
and most importantly the ability to
and most importantly the ability to learn that would be huge so let’s think
learn that would be huge so let’s think
learn that would be huge so let’s think a little bit what the brain can really
a little bit what the brain can really
a little bit what the brain can really do which makes it so amazing and I
do which makes it so amazing and I
do which makes it so amazing and I selected the rather trivial thing here
selected the rather trivial thing here
selected the rather trivial thing here which is just playing tennis and we said
which is just playing tennis and we said
which is just playing tennis and we said ok like also robot can do but it’s not
ok like also robot can do but it’s not
ok like also robot can do but it’s not really true because if you think what we
really true because if you think what we
really true because if you think what we do here is quite impressive if you are
do here is quite impressive if you are
do here is quite impressive if you are playing tennis you can play tennis and
playing tennis you can play tennis and
playing tennis you can play tennis and any tennis called in the world even if
any tennis called in the world even if
any tennis called in the world even if you have never seen that particular
you have never seen that particular
you have never seen that particular before so you immediately adopted the
before so you immediately adopted the
before so you immediately adopted the environment based on the learning you
environment based on the learning you
environment based on the learning you perform many many years before it took
perform many many years before it took
perform many many years before it took you a while to get that far but once you
you a while to get that far but once you
you a while to get that far but once you are there you can do all these things
are there you can do all these things
are there you can do all these things then there’s a ball coming towards your
then there’s a ball coming towards your
then there’s a ball coming towards your tennis ball it was really high speed and
tennis ball it was really high speed and
tennis ball it was really high speed and you see it and you can move your arm and
you see it and you can move your arm and
you see it and you can move your arm and you can make something which is very
you can make something which is very
you can make something which is very important based on the input from your
important based on the input from your
important based on the input from your sensors from your vision sensors also
sensors from your vision sensors also
sensors from your vision sensors also from your appropriate sensors that that
from your appropriate sensors that that
from your appropriate sensors that that measure the position of your arms and
measure the position of your arms and
measure the position of your arms and your leg you can make a prediction you
your leg you can make a prediction you
your leg you can make a prediction you can make a prediction where and when the
can make a prediction where and when the
can make a prediction where and when the ball will hit your racket okay and that
ball will hit your racket okay and that
ball will hit your racket okay and that is prediction making based on very
is prediction making based on very
is prediction making based on very complex data entering your brain then
complex data entering your brain then
complex data entering your brain then you can hit the ball okay and can move
you can hit the ball okay and can move
you can hit the ball okay and can move your muscles in a controlled way and you
your muscles in a controlled way and you
your muscles in a controlled way and you want that the ball is not caught by your
want that the ball is not caught by your
want that the ball is not caught by your opponent so you sent to sent it
opponent so you sent to sent it
opponent so you sent to sent it somewhere else again it asks you to make
somewhere else again it asks you to make
somewhere else again it asks you to make a prediction
a prediction
a prediction so using complex data for prediction
so using complex data for prediction
so using complex data for prediction making is probably the single most
making is probably the single most
making is probably the single most important thing that our brain can do
important thing that our brain can do
important thing that our brain can do and we have a hard time doing this for
and we have a hard time doing this for
and we have a hard time doing this for computers so building a computer like
computers so building a computer like
computers so building a computer like the brain for these kind of things and
the brain for these kind of things and
the brain for these kind of things and I’m not only talking about robotics
I’m not only talking about robotics
I’m not only talking about robotics complex data can be more abstract it can
complex data can be more abstract it can
complex data can be more abstract it can be industry data for example business
be industry data for example business
be industry data for example business data science data medical data many kind
data science data medical data many kind
data science data medical data many kind of things look into the time sequence
of things look into the time sequence
of things look into the time sequence and make predictions would be great okay
and make predictions would be great okay
and make predictions would be great okay and so let’s have a little look on what
and so let’s have a little look on what
and so let’s have a little look on what these systems that we use all the time
these systems that we use all the time
these systems that we use all the time look like microscopically and although
look like microscopically and although
look like microscopically and although this is not quite the same scale it so
this is not quite the same scale it so
this is not quite the same scale it so tells you that these things are
tells you that these things are
tells you that these things are different on the right you see a
different on the right you see a
different on the right you see a microprocessor an Intel processor and
microprocessor an Intel processor and
microprocessor an Intel processor and you see it has these kind of a
you see it has these kind of a
you see it has these kind of a necessities very positively engineering
necessities very positively engineering
necessities very positively engineering approach it has clearly clearly defined
approach it has clearly clearly defined
approach it has clearly clearly defined components like memory routers
components like memory routers
components like memory routers connections compute units that are
connections compute units that are
connections compute units that are connected to form a computer
connected to form a computer
connected to form a computer architecture that has been engineered
architecture that has been engineered
architecture that has been engineered and that is mostly based on the idea of
and that is mostly based on the idea of
and that is mostly based on the idea of John for Norman and others we call that
John for Norman and others we call that
John for Norman and others we call that the phenomena architecture on the left
the phenomena architecture on the left
the phenomena architecture on the left you see sort of a millimeter slice of
you see sort of a millimeter slice of
you see sort of a millimeter slice of cortex of you
cortex of you
cortex of you cortex it’s a reconstructed one it’s not
cortex it’s a reconstructed one it’s not
cortex it’s a reconstructed one it’s not a real one and you see it has this it
a real one and you see it has this it
a real one and you see it has this it looks different it looks kind of regular
looks different it looks kind of regular
looks different it looks kind of regular rather uniform it has a layered type
rather uniform it has a layered type
rather uniform it has a layered type structure but if you would keep plotting
structure but if you would keep plotting
structure but if you would keep plotting these things over a large distance it
these things over a large distance it
these things over a large distance it would look like like everywhere the same
would look like like everywhere the same
would look like like everywhere the same and what it has is these little yellow
and what it has is these little yellow
and what it has is these little yellow dots they are yellow on this
dots they are yellow on this
dots they are yellow on this reconstruction and those are neural
reconstruction and those are neural
reconstruction and those are neural cells and there are fibers connecting
cells and there are fibers connecting
cells and there are fibers connecting these neural cells mostly vertically but
these neural cells mostly vertically but
these neural cells mostly vertically but sometimes also horizontally with each
sometimes also horizontally with each
sometimes also horizontally with each other and this is the system that does
other and this is the system that does
other and this is the system that does this amazing information processing that
this amazing information processing that
this amazing information processing that I mentioned before so how can we handle
I mentioned before so how can we handle
I mentioned before so how can we handle this how can we build a system that
this how can we build a system that
this how can we build a system that works like the brain where one is one
works like the brain where one is one
works like the brain where one is one idea is to take the left thing and put
idea is to take the left thing and put
idea is to take the left thing and put it on the right thing in other words to
it on the right thing in other words to
it on the right thing in other words to simulate okay and all right what you see
simulate okay and all right what you see
simulate okay and all right what you see here it’s hard to see it’s kind of
here it’s hard to see it’s kind of
here it’s hard to see it’s kind of darkish blue and it’s work done by Henry
darkish blue and it’s work done by Henry
darkish blue and it’s work done by Henry Markram
Markram
Markram one of my friends yet the human brain
one of my friends yet the human brain
one of my friends yet the human brain project and these people simulate brain
project and these people simulate brain
project and these people simulate brain cells on supercomputers and the
cells on supercomputers and the
cells on supercomputers and the simulation the important thing about
simulation the important thing about
simulation the important thing about simulation is that simulation tells you
simulation is that simulation tells you
simulation is that simulation tells you how things change with time simulations
how things change with time simulations
how things change with time simulations are not just static images but they tell
are not just static images but they tell
are not just static images but they tell you how things change with time and this
you how things change with time and this
you how things change with time and this is what Henry does and if he runs the
is what Henry does and if he runs the
is what Henry does and if he runs the simulation these are about 10,000
simulation these are about 10,000
simulation these are about 10,000 neurons 10 to the 4 they’re about 10 to
neurons 10 to the 4 they’re about 10 to
neurons 10 to the 4 they’re about 10 to the 11 and our brains are this is a
the 11 and our brains are this is a
the 11 and our brains are this is a small system you see that they fire so
small system you see that they fire so
small system you see that they fire so they communicate by electrical pulses
they communicate by electrical pulses
they communicate by electrical pulses and the colors here represent the
and the colors here represent the
and the colors here represent the voltage between the inside and the
voltage between the inside and the
voltage between the inside and the outside of the cell and if you analyze
outside of the cell and if you analyze
outside of the cell and if you analyze this data you find well it looks a lot
this data you find well it looks a lot
this data you find well it looks a lot like biological systems but there are
like biological systems but there are
like biological systems but there are problems one problem is that this
problems one problem is that this
problems one problem is that this simulation takes a lot of energy and if
simulation takes a lot of energy and if
simulation takes a lot of energy and if you look at the second row here energy
you look at the second row here energy
you look at the second row here energy per operation what do I mean by
per operation what do I mean by
per operation what do I mean by operation but it doesn’t really matter
operation but it doesn’t really matter
operation but it doesn’t really matter but it could be like producing one of
but it could be like producing one of
but it could be like producing one of these spikes which you have just seen so
these spikes which you have just seen so
these spikes which you have just seen so if you count the zeros there there are
if you count the zeros there there are
if you count the zeros there there are 14 so the simulation uses
14 so the simulation uses
14 so the simulation uses 10 to the 14 times more energy than the
10 to the 14 times more energy than the
10 to the 14 times more energy than the biological example 10 to the 14 is 10 by
biological example 10 to the 14 is 10 by
biological example 10 to the 14 is 10 by 10 by 10 by 10 by 10 by 10 by 10 by 10
10 by 10 by 10 by 10 by 10 by 10 by 10
10 by 10 by 10 by 10 by 10 by 10 by 10 by 10 by 10 by 10 by 10 by 10 by 10 that
by 10 by 10 by 10 by 10 by 10 by 10 that
by 10 by 10 by 10 by 10 by 10 by 10 that is a lot okay and that’s not all there
is a lot okay and that’s not all there
is a lot okay and that’s not all there is another problem and this is a most
is another problem and this is a most
is another problem and this is a most serious problem is that the simulation
serious problem is that the simulation
serious problem is that the simulation although it looks very fleshy but of
although it looks very fleshy but of
although it looks very fleshy but of course that’s just a movie
course that’s just a movie
course that’s just a movie and it doesn’t tell you the real
and it doesn’t tell you the real
and it doesn’t tell you the real simulation speed the simulation is a
simulation speed the simulation is a
simulation speed the simulation is a thousand times slower than biology is
thousand times slower than biology is
thousand times slower than biology is that a problem well I just said that to
that a problem well I just said that to
that a problem well I just said that to learn tennis or to learn languages or to
learn tennis or to learn languages or to
learn tennis or to learn languages or to learn whatever we do to develop to build
learn whatever we do to develop to build
learn whatever we do to develop to build an internal model of our world in our
an internal model of our world in our
an internal model of our world in our brain we need typically a year so this
brain we need typically a year so this
brain we need typically a year so this sentence is wrong there I’m sorry it
sentence is wrong there I’m sorry it
sentence is wrong there I’m sorry it should say a day of learning or say a
should say a day of learning or say a
should say a day of learning or say a year of learning needs about a thousand
year of learning needs about a thousand
year of learning needs about a thousand years of simulation which is totally out
years of simulation which is totally out
years of simulation which is totally out of range there is no way you can run a
of range there is no way you can run a
of range there is no way you can run a supercomputer for thousands of years
supercomputer for thousands of years
supercomputer for thousands of years and at the same time if you would scale
and at the same time if you would scale
and at the same time if you would scale it up to the human brain the simulation
it up to the human brain the simulation
it up to the human brain the simulation would use about five times all the
would use about five times all the
would use about five times all the electrical power that is being produced
electrical power that is being produced
electrical power that is being produced in Germany which you would have to focus
in Germany which you would have to focus
in Germany which you would have to focus on one computer you have to run
on one computer you have to run
on one computer you have to run thousands of years so it’s just no way
thousands of years so it’s just no way
thousands of years so it’s just no way that you can do that now let’s look a
that you can do that now let’s look a
that you can do that now let’s look a little bit this is this is a little bit
little bit this is this is a little bit
little bit this is this is a little bit about mathematics also okay I’m sorry
about mathematics also okay I’m sorry
about mathematics also okay I’m sorry this is an equation but this is a
this is an equation but this is a
this is an equation but this is a Technical University sorry this is the
Technical University sorry this is the
Technical University sorry this is the Technical University so you know what an
Technical University so you know what an
Technical University so you know what an equation is okay is it horrible this is
equation is okay is it horrible this is
equation is okay is it horrible this is a differential equation and a
a differential equation and a
a differential equation and a differential equation the most important
differential equation the most important
differential equation the most important thing on the left side at the bottom
thing on the left side at the bottom
thing on the left side at the bottom there you see in the equation you see a
there you see in the equation you see a
there you see in the equation you see a t a little T and the T is time okay and
t a little T and the T is time okay and
t a little T and the T is time okay and then you see a big V and that’s voltage
then you see a big V and that’s voltage
then you see a big V and that’s voltage so this equation tells you how the
so this equation tells you how the
so this equation tells you how the voltage between the inside and the
voltage between the inside and the
voltage between the inside and the outside of the cell changes its function
outside of the cell changes its function
outside of the cell changes its function of time and this is what hey
of time and this is what hey
of time and this is what hey reuses for simulations so if you go to a
reuses for simulations so if you go to a
reuses for simulations so if you go to a supercomputer what you do is you solve
supercomputer what you do is you solve
supercomputer what you do is you solve these equations and you solve tens of
these equations and you solve tens of
these equations and you solve tens of thousands of them and they are connected
thousands of them and they are connected
thousands of them and they are connected by these summations which you see there
by these summations which you see there
by these summations which you see there now this is why it takes so long because
now this is why it takes so long because
now this is why it takes so long because those who know a little bit about
those who know a little bit about
those who know a little bit about mathematics a tiny little bit you see
mathematics a tiny little bit you see
mathematics a tiny little bit you see that the derivative is proportional to
that the derivative is proportional to
that the derivative is proportional to the voltage is there a function where
the voltage is there a function where
the voltage is there a function where the derivative is proportional to the
the derivative is proportional to the
the derivative is proportional to the voltage I guess you probably do that
voltage I guess you probably do that
voltage I guess you probably do that kind of thing at school it’s called the
kind of thing at school it’s called the
kind of thing at school it’s called the exponential function so you have to
exponential function so you have to
exponential function so you have to calculate lots of exponential functions
calculate lots of exponential functions
calculate lots of exponential functions exponential functions look like that
exponential functions look like that
exponential functions look like that they are not linear and it takes a lot
they are not linear and it takes a lot
they are not linear and it takes a lot of time to calculate them now you have
of time to calculate them now you have
of time to calculate them now you have to move data across your phenomen
to move data across your phenomen
to move data across your phenomen machine all the time why this is why
machine all the time why this is why
machine all the time why this is why these simulations are so horribly
these simulations are so horribly
these simulations are so horribly inefficient now at the bottom there is
inefficient now at the bottom there is
inefficient now at the bottom there is anyway what we really have we know that
anyway what we really have we know that
anyway what we really have we know that our selves are can be n is to some
our selves are can be n is to some
our selves are can be n is to some extent replaced by electrical circuits
extent replaced by electrical circuits
extent replaced by electrical circuits like batteries capacitors and and
like batteries capacitors and and
like batteries capacitors and and conductances or resistors so what if we
conductances or resistors so what if we
conductances or resistors so what if we do something crazy it was a big if out
do something crazy it was a big if out
do something crazy it was a big if out we give up computing we don’t compute
we give up computing we don’t compute
we give up computing we don’t compute anymore we build computers that don’t
anymore we build computers that don’t
anymore we build computers that don’t compute that just our physical copies of
compute that just our physical copies of
compute that just our physical copies of the brain the brain moves ions through
the brain the brain moves ions through
the brain the brain moves ions through ourselves and we move electrons through
ourselves and we move electrons through
ourselves and we move electrons through transistors how about that ok this is a
transistors how about that ok this is a
transistors how about that ok this is a transistor you probably have seen these
transistor you probably have seen these
transistor you probably have seen these things transistors have three legs okay
things transistors have three legs okay
things transistors have three legs okay and these are the blue things
and these are the blue things
and these are the blue things the legs are source and drain and gate
the legs are source and drain and gate
the legs are source and drain and gate and this is a computer drawing of a
and this is a computer drawing of a
and this is a computer drawing of a transistor it’s not a real one it’s
transistor it’s not a real one it’s
transistor it’s not a real one it’s viewed from the top and the thing the
viewed from the top and the thing the
viewed from the top and the thing the blue thing on top is called the gate and
blue thing on top is called the gate and
blue thing on top is called the gate and the length of the gate
the length of the gate
the length of the gate somehow characterizes the size of the
somehow characterizes the size of the
somehow characterizes the size of the transistors this is in real life
transistors this is in real life
transistors this is in real life typically today around 10 nanometers or
typically today around 10 nanometers or
typically today around 10 nanometers or something like that now if you want to
something like that now if you want to
something like that now if you want to build a copy of the brain you have to be
build a copy of the brain you have to be
build a copy of the brain you have to be clever and wire up a circuit that works
clever and wire up a circuit that works
clever and wire up a circuit that works exactly like a brain works so this is
exactly like a brain works so this is
exactly like a brain works so this is all you have or you have a transistors
all you have or you have a transistors
all you have or you have a transistors and now you sit on your computer and you
and now you sit on your computer and you
and now you sit on your computer and you wire them up
wire them up
wire them up now we wire them up okay you see many
now we wire them up okay you see many
now we wire them up okay you see many transistors now like tens of transistors
transistors now like tens of transistors
transistors now like tens of transistors now you see maybe a hundred transistors
now you see maybe a hundred transistors
now you see maybe a hundred transistors and they look rather irregular we zoom
and they look rather irregular we zoom
and they look rather irregular we zoom out and we see that the structure starts
out and we see that the structure starts
out and we see that the structure starts to be more and more regular and what you
to be more and more regular and what you
to be more and more regular and what you see here are circuits for synapses you
see here are circuits for synapses you
see here are circuits for synapses you now see about I don’t know a thousand
now see about I don’t know a thousand
now see about I don’t know a thousand synapses we move out more and more now
synapses we move out more and more now
synapses we move out more and more now we see about 50,000 synapses and we see
we see about 50,000 synapses and we see
we see about 50,000 synapses and we see the border of what’s a chip five
the border of what’s a chip five
the border of what’s a chip five millimeter my times 10 me a hundred
millimeter my times 10 me a hundred
millimeter my times 10 me a hundred million ten millimeter then we see many
million ten millimeter then we see many
million ten millimeter then we see many chips all right there are also neurons
chips all right there are also neurons
chips all right there are also neurons on there and all these chips sit on
on there and all these chips sit on
on there and all these chips sit on something which we call a wafer that’s a
something which we call a wafer that’s a
something which we call a wafer that’s a disc of silicon and on this disc of
disc of silicon and on this disc of
disc of silicon and on this disc of silicon we have two hundred thousand
silicon we have two hundred thousand
silicon we have two hundred thousand neurons physical models they don’t solve
neurons physical models they don’t solve
neurons physical models they don’t solve equations but they are just circuits
equations but they are just circuits
equations but they are just circuits like the brain and 250 million synapses
like the brain and 250 million synapses
like the brain and 250 million synapses which can learn reconfigure and do all
which can learn reconfigure and do all
which can learn reconfigure and do all kinds of cool things and the great thing
kinds of cool things and the great thing
kinds of cool things and the great thing is you remember the circuit there was a
is you remember the circuit there was a
is you remember the circuit there was a resistor and a capacitor and this is an
resistor and a capacitor and this is an
resistor and a capacitor and this is an electrical engineering school also you
electrical engineering school also you
electrical engineering school also you know what happens if you multiply the
know what happens if you multiply the
know what happens if you multiply the value of a resistor and the capacitor
value of a resistor and the capacitor
value of a resistor and the capacitor you get a time and that’s the speed of
you get a time and that’s the speed of
you get a time and that’s the speed of the system
the system
the system it’s the decay time of an exponential
it’s the decay time of an exponential
it’s the decay time of an exponential that tells you how fast your system runs
that tells you how fast your system runs
that tells you how fast your system runs so if you are clever you could build a
so if you are clever you could build a
so if you are clever you could build a system where you scale you scale all the
system where you scale you scale all the
system where you scale you scale all the time constant by same factor say ten
time constant by same factor say ten
time constant by same factor say ten thousand so you could build a system
thousand so you could build a system
thousand so you could build a system like this one here which runs ten
like this one here which runs ten
like this one here which runs ten thousand times faster than biology why
thousand times faster than biology why
thousand times faster than biology why would you do that well think of the day
would you do that well think of the day
would you do that well think of the day of learning which in real life takes a
of learning which in real life takes a
of learning which in real life takes a day if you run it on a supercomputer it
day if you run it on a supercomputer it
day if you run it on a supercomputer it takes a thousand days if you run it on
takes a thousand days if you run it on
takes a thousand days if you run it on this system
this system
this system it runs ten seconds and that is the
it runs ten seconds and that is the
it runs ten seconds and that is the system to study learning and development
system to study learning and development
system to study learning and development which is really the key to making
which is really the key to making
which is really the key to making artificial neural systems work that was
artificial neural systems work that was
artificial neural systems work that was a computer drawing what you see here is
a computer drawing what you see here is
a computer drawing what you see here is real life this is a real piece of
real life this is a real piece of
real life this is a real piece of silicon ok
silicon ok
silicon ok it’s a
it’s a
it’s a many many chips you see them these
many many chips you see them these
many many chips you see them these little things which you saw on the
little things which you saw on the
little things which you saw on the computer drawing and this is a disc of
computer drawing and this is a disc of
computer drawing and this is a disc of 20 centimeters diameter and it contains
20 centimeters diameter and it contains
20 centimeters diameter and it contains as I said 50 million synapses end for
as I said 50 million synapses end for
as I said 50 million synapses end for end 200,000 neurons which you can wire
end 200,000 neurons which you can wire
end 200,000 neurons which you can wire up any way you like
up any way you like
up any way you like now of course having a disc of silicon
now of course having a disc of silicon
now of course having a disc of silicon is nothing you have to build a system
is nothing you have to build a system
is nothing you have to build a system out of it and I am tempted to call it a
out of it and I am tempted to call it a
out of it and I am tempted to call it a computer but it’s not really a computer
computer but it’s not really a computer
computer but it’s not really a computer it’s a physical model of a brain circuit
it’s a physical model of a brain circuit
it’s a physical model of a brain circuit this is one of those units which looks
this is one of those units which looks
this is one of those units which looks very strange it’s about that big and you
very strange it’s about that big and you
very strange it’s about that big and you see a lot of conventional electronic
see a lot of conventional electronic
see a lot of conventional electronic because you have to send power into the
because you have to send power into the
because you have to send power into the system you have to send data into the
system you have to send data into the
system you have to send data into the system you have to read out the data but
system you have to read out the data but
system you have to read out the data but underneath this pile of electronics is
underneath this pile of electronics is
underneath this pile of electronics is the matching by wafer which we call a
the matching by wafer which we call a
the matching by wafer which we call a neuromorphic system the banana finally
neuromorphic system the banana finally
neuromorphic system the banana finally there’s a real reason for it that’s at
there’s a real reason for it that’s at
there’s a real reason for it that’s at all pay off should we really do this if
all pay off should we really do this if
all pay off should we really do this if you eat a banana per hour okay if you
you eat a banana per hour okay if you
you eat a banana per hour okay if you just eat one banana you get energy
just eat one banana you get energy
just eat one banana you get energy energy you measure in kilo calories or
energy you measure in kilo calories or
energy you measure in kilo calories or Joule if you eat one of these things you
Joule if you eat one of these things you
Joule if you eat one of these things you get for your body and your brain 500,000
get for your body and your brain 500,000
get for your body and your brain 500,000 Joule if you eat one banana per hour
Joule if you eat one banana per hour
Joule if you eat one banana per hour that’s about 140 watts or this is a
that’s about 140 watts or this is a
that’s about 140 watts or this is a little bit of their 0.2 horsepowers
little bit of their 0.2 horsepowers
little bit of their 0.2 horsepowers which is enough to run your body you can
which is enough to run your body you can
which is enough to run your body you can live from a banana per hour
live from a banana per hour
live from a banana per hour I’m not saying is very healthy but
I’m not saying is very healthy but
I’m not saying is very healthy but energy wise energy wise that’s
energy wise energy wise that’s
energy wise energy wise that’s sufficient so how does this look like
sufficient so how does this look like
sufficient so how does this look like now in terms of computing the energy
now in terms of computing the energy
now in terms of computing the energy needed for sales synaptic transmission
needed for sales synaptic transmission
needed for sales synaptic transmission you know neurons are connected by
you know neurons are connected by
you know neurons are connected by synapses and we transmit a signal it
synapses and we transmit a signal it
synapses and we transmit a signal it requires a certain energy and if you
requires a certain energy and if you
requires a certain energy and if you look to our own brain and I’m not
look to our own brain and I’m not
look to our own brain and I’m not repeating all the zeros now this is now
repeating all the zeros now this is now
repeating all the zeros now this is now of course behind the zero you need about
of course behind the zero you need about
of course behind the zero you need about 10 to the minus 14 jul 2015 10
which is really very very efficient if
which is really very very efficient if you run the simulation on a
you run the simulation on a
you run the simulation on a supercomputer like the ones you have
supercomputer like the ones you have
supercomputer like the ones you have here in hewlett for example not far from
here in hewlett for example not far from
here in hewlett for example not far from here you typically use up to 1 joule so
here you typically use up to 1 joule so
here you typically use up to 1 joule so you only get 500,000 synaptic
you only get 500,000 synaptic
you only get 500,000 synaptic transmissions on a computer if you burn
transmissions on a computer if you burn
transmissions on a computer if you burn a banana and run a power station from it
a banana and run a power station from it
a banana and run a power station from it ok
ok
ok and your amorphic systems are somehow in
and your amorphic systems are somehow in
and your amorphic systems are somehow in between but on a very favorable way the
between but on a very favorable way the
between but on a very favorable way the neuromorphic systems are typically
neuromorphic systems are typically
neuromorphic systems are typically around 10 to the minus 10 joules so they
around 10 to the minus 10 joules so they
around 10 to the minus 10 joules so they are only a factor 10,000 away from
are only a factor 10,000 away from
are only a factor 10,000 away from biology and the factor 10 to the 10 away
biology and the factor 10 to the 10 away
biology and the factor 10 to the 10 away from conventional supercomputers so
from conventional supercomputers so
from conventional supercomputers so these are very energy efficient brain
these are very energy efficient brain
these are very energy efficient brain inspired computing systems more
inspired computing systems more
inspired computing systems more important is and that’s a bit difficult
important is and that’s a bit difficult
important is and that’s a bit difficult to understand time for a day or it’s
to understand time for a day or it’s
to understand time for a day or it’s time for a day a day takes a day well it
time for a day a day takes a day well it
time for a day a day takes a day well it doesn’t take a day if you look to your
doesn’t take a day if you look to your
doesn’t take a day if you look to your brain a day is a day ok that’s what we
brain a day is a day ok that’s what we
brain a day is a day ok that’s what we call real time again if you go to a
call real time again if you go to a
call real time again if you go to a supercomputer a day takes a thousand
supercomputer a day takes a thousand
supercomputer a day takes a thousand days because of the slowdown factor on
days because of the slowdown factor on
days because of the slowdown factor on the neuromorphic site systems the one we
the neuromorphic site systems the one we
the neuromorphic site systems the one we call brain scales and our lab at home
call brain scales and our lab at home
call brain scales and our lab at home it’s about 10 seconds so we have a
it’s about 10 seconds so we have a
it’s about 10 seconds so we have a system that is fault tolerant it runs
system that is fault tolerant it runs
system that is fault tolerant it runs even faster than biology and it’s
even faster than biology and it’s
even faster than biology and it’s extremely energy efficient and now we
extremely energy efficient and now we
extremely energy efficient and now we can go and do what people do with neural
can go and do what people do with neural
can go and do what people do with neural networks now like the deep learning
networks now like the deep learning
networks now like the deep learning experiments which are still very slow in
experiments which are still very slow in
experiments which are still very slow in learning and very energy inefficient and
learning and very energy inefficient and
learning and very energy inefficient and so what we do now is to move all the
so what we do now is to move all the
so what we do now is to move all the knowledge that we have on your networks
knowledge that we have on your networks
knowledge that we have on your networks and we add our knowledge of learning to
and we add our knowledge of learning to
and we add our knowledge of learning to make fast energy efficient learning
make fast energy efficient learning
make fast energy efficient learning systems to solve problems that we have
systems to solve problems that we have
systems to solve problems that we have to solve for cognitive computing and the
to solve for cognitive computing and the
to solve for cognitive computing and the problems are typically those which use
problems are typically those which use
problems are typically those which use complex data find the structure and
complex data find the structure and
complex data find the structure and complex data and make predictions in
complex data and make predictions in
complex data and make predictions in time and as I said the data can come
time and as I said the data can come
time and as I said the data can come from many sources it can come from
from many sources it can come from
from many sources it can come from sensors vision sensors movies things
sensors vision sensors movies things
sensors vision sensors movies things like that but also more abstract data
like that but also more abstract data
like that but also more abstract data like business data and in particular
like business data and in particular
like business data and in particular science and medical data and this is
science and medical data and this is
science and medical data and this is what we think is a very important
what we think is a very important
what we think is a very important contribution to make cognitive computing
contribution to make cognitive computing
contribution to make cognitive computing a reality
a reality
a reality now this was very superficial of course
now this was very superficial of course
now this was very superficial of course I just get 18 minutes and they are over
I just get 18 minutes and they are over
I just get 18 minutes and they are over now if you want to know more and I’m
now if you want to know more and I’m
now if you want to know more and I’m sure you want to know more if you even
sure you want to know more if you even
sure you want to know more if you even want to try it yourself I’m not allowed
want to try it yourself I’m not allowed
want to try it yourself I’m not allowed to show you’re else here people told me
to show you’re else here people told me
to show you’re else here people told me no propaganda so I do the propaganda by
no propaganda so I do the propaganda by
no propaganda so I do the propaganda by telling you there is a website where you
telling you there is a website where you
telling you there is a website where you can see all the details and it’s very
can see all the details and it’s very
can see all the details and it’s very easy to remember
easy to remember
easy to remember it’s called neuromorphic dot EU thank
it’s called neuromorphic dot EU thank
it’s called neuromorphic dot EU thank you very much
”
Be First to Comment