let me start by telling you that for me air is pretty much the same as as water
air is pretty much the same as as water
air is pretty much the same as as water now hearing that you probably think I’m
now hearing that you probably think I’m
now hearing that you probably think I’m a bit weird because there are some
a bit weird because there are some
a bit weird because there are some obvious differences say if you see this
obvious differences say if you see this
obvious differences say if you see this picture of a glass of water and you
picture of a glass of water and you
picture of a glass of water and you consider the air in this room they’re
consider the air in this room they’re
consider the air in this room they’re obviously very different and luckily the
obviously very different and luckily the
obviously very different and luckily the room for example is not filled with
room for example is not filled with
room for example is not filled with water otherwise we’d all drown so what I
water otherwise we’d all drown so what I
water otherwise we’d all drown so what I mean with being the same is physics wise
mean with being the same is physics wise
mean with being the same is physics wise you can actually look at them in pretty
you can actually look at them in pretty
you can actually look at them in pretty much the same way and that becomes more
much the same way and that becomes more
much the same way and that becomes more apparent if you have something to show
apparent if you have something to show
apparent if you have something to show how they’re moving so here you can
how they’re moving so here you can
how they’re moving so here you can actually see it’s just a puff of smoke
actually see it’s just a puff of smoke
actually see it’s just a puff of smoke actually and if you look at the shapes
actually and if you look at the shapes
actually and if you look at the shapes that form if you have something like ink
that form if you have something like ink
that form if you have something like ink and water then they’re actually very
and water then they’re actually very
and water then they’re actually very similar as you can see here and physics
similar as you can see here and physics
similar as you can see here and physics wise
wise
wise interestingly also the same kind of
interestingly also the same kind of
interestingly also the same kind of phenomena are driving these shapes and
phenomena are driving these shapes and
phenomena are driving these shapes and motions and that typically summarized as
motions and that typically summarized as
motions and that typically summarized as turbulence so the reason I bring that up
turbulence so the reason I bring that up
turbulence so the reason I bring that up here is because turbulence is actually
here is because turbulence is actually
here is because turbulence is actually still one of the big remaining mysteries
still one of the big remaining mysteries
still one of the big remaining mysteries of modern day physics and is actually so
of modern day physics and is actually so
of modern day physics and is actually so fundamental that Institute’s like this
fundamental that Institute’s like this
fundamental that Institute’s like this clay mathematics Institute are actually
clay mathematics Institute are actually
clay mathematics Institute are actually paying a million dollars for anyone who
paying a million dollars for anyone who
paying a million dollars for anyone who figures it out so if you’re bored
figures it out so if you’re bored
figures it out so if you’re bored tomorrow afternoon you might want to
tomorrow afternoon you might want to
tomorrow afternoon you might want to give this a thought mmm now you’re
give this a thought mmm now you’re
give this a thought mmm now you’re probably wondering why anyone would pay
probably wondering why anyone would pay
probably wondering why anyone would pay a million dollars for a problem like
a million dollars for a problem like
a million dollars for a problem like that and the reason is that it’s really
that and the reason is that it’s really
that and the reason is that it’s really fundamental so starting with the air you
fundamental so starting with the air you
fundamental so starting with the air you breathe with the plane you take to an
breathe with the plane you take to an
breathe with the plane you take to an explication
explication
explication or also think you think about what’s
or also think you think about what’s
or also think you think about what’s happening in your body the blood flowing
happening in your body the blood flowing
happening in your body the blood flowing through your veins they actually all
through your veins they actually all
through your veins they actually all behave in accordance to these underlying
behave in accordance to these underlying
behave in accordance to these underlying physics and despite looking very
physics and despite looking very
physics and despite looking very different we can actually look physics
different we can actually look physics
different we can actually look physics wise look at these materials together
wise look at these materials together
wise look at these materials together and they typically summarized as fluids
and they typically summarized as fluids
and they typically summarized as fluids so that’s what I’ll also use in the
so that’s what I’ll also use in the
so that’s what I’ll also use in the following for these different different
following for these different different
following for these different different materials now I personally look at those
materials now I personally look at those
materials now I personally look at those formal computer graphics perspective so
formal computer graphics perspective so
formal computer graphics perspective so to show some examples this is one of our
to show some examples this is one of our
to show some examples this is one of our older works
older works
older works a while back where we simulator to a
a while back where we simulator to a
a while back where we simulator to a kind of burning car at the trail of
kind of burning car at the trail of
kind of burning car at the trail of smoke behind it
smoke behind it
smoke behind it and
and
and the four million particles that you can
the four million particles that you can
the four million particles that you can see in the top right corner there
see in the top right corner there
see in the top right corner there indicates a bit that there’s a lot of
indicates a bit that there’s a lot of
indicates a bit that there’s a lot of work going into into producing that
work going into into producing that
work going into into producing that animation here or another example is
animation here or another example is
animation here or another example is this truck being chased by a dark cloud
this truck being chased by a dark cloud
this truck being chased by a dark cloud of smoke that you can see here so
of smoke that you can see here so
of smoke that you can see here so overall summary here is these fluids are
overall summary here is these fluids are
overall summary here is these fluids are really important and popping up in many
really important and popping up in many
really important and popping up in many many situations so how do we go about
many situations so how do we go about
many situations so how do we go about kind of working with them or analyzing
kind of working with them or analyzing
kind of working with them or analyzing them and for that we can now look at
them and for that we can now look at
them and for that we can now look at kind of the principles of science and if
kind of the principles of science and if
kind of the principles of science and if you think about how that’s been
you think about how that’s been
you think about how that’s been happening way back for example with
happening way back for example with
happening way back for example with Newton he was as the story goes sitting
Newton he was as the story goes sitting
Newton he was as the story goes sitting on the tree an apple falling down onto
on the tree an apple falling down onto
on the tree an apple falling down onto his head or beside him is not fully
his head or beside him is not fully
his head or beside him is not fully confirmed but in a way that was his
confirmed but in a way that was his
confirmed but in a way that was his observation his data and from that he
observation his data and from that he
observation his data and from that he developed these really condensed and and
developed these really condensed and and
developed these really condensed and and nice equations the model and after which
nice equations the model and after which
nice equations the model and after which with which to model reality here so
with which to model reality here so
with which to model reality here so those are two really fundamental topics
those are two really fundamental topics
those are two really fundamental topics in science experiment and theory and
in science experiment and theory and
in science experiment and theory and coming back to these fluids they’ve
coming back to these fluids they’ve
coming back to these fluids they’ve actually been studied even before Newton
actually been studied even before Newton
actually been studied even before Newton was around so for example this is one of
was around so for example this is one of
was around so for example this is one of the famous sketches of Leonardo da Vinci
the famous sketches of Leonardo da Vinci
the famous sketches of Leonardo da Vinci we analyzed the or studied the motions
we analyzed the or studied the motions
we analyzed the or studied the motions of water and this world’s forming and
of water and this world’s forming and
of water and this world’s forming and especially for these days actually a
especially for these days actually a
especially for these days actually a third component that’s really important
third component that’s really important
third component that’s really important those are computations so with computers
those are computations so with computers
those are computations so with computers we can now if we solve very complicated
we can now if we solve very complicated
we can now if we solve very complicated models and predict what’s happening in
models and predict what’s happening in
models and predict what’s happening in the nature now the reason I’m standing
the nature now the reason I’m standing
the nature now the reason I’m standing here today is that we have a kind of new
here today is that we have a kind of new
here today is that we have a kind of new kid in town these deep learning
kid in town these deep learning
kid in town these deep learning techniques and I know that’s quite a
techniques and I know that’s quite a
techniques and I know that’s quite a password by now so before he heard it
password by now so before he heard it
password by now so before he heard it too many times recently in the news and
too many times recently in the news and
too many times recently in the news and every we’re talking about buzzwords
every we’re talking about buzzwords
every we’re talking about buzzwords here’s another one but I can promise
here’s another one but I can promise
here’s another one but I can promise that I won’t be talking about block
that I won’t be talking about block
that I won’t be talking about block chains at all so I’ll focus on this deep
chains at all so I’ll focus on this deep
chains at all so I’ll focus on this deep learning and the following and in order
learning and the following and in order
learning and the following and in order to understand that I also want to
to understand that I also want to
to understand that I also want to explain a bit how that works because
explain a bit how that works because
explain a bit how that works because what is actually driving pretty much all
what is actually driving pretty much all
what is actually driving pretty much all of the techniques you you find
of the techniques you you find
of the techniques you you find summarized as deep learning are what’s
summarized as deep learning are what’s
summarized as deep learning are what’s called neural net
called neural net
called neural net so those are roughly modeled by our
so those are roughly modeled by our
so those are roughly modeled by our probably quite limited understanding of
probably quite limited understanding of
probably quite limited understanding of the human brain but as the name suggests
the human brain but as the name suggests
the human brain but as the name suggests it’s a network of nodes so I want to
it’s a network of nodes so I want to
it’s a network of nodes so I want to give an example here this blue dot that
give an example here this blue dot that
give an example here this blue dot that you can see here might not now be one of
you can see here might not now be one of
you can see here might not now be one of these nodes in the network and we now
these nodes in the network and we now
these nodes in the network and we now have some input coming in this example
have some input coming in this example
have some input coming in this example from the from the left going out on the
from the from the left going out on the
from the from the left going out on the right each of these arrows is actually
right each of these arrows is actually
right each of these arrows is actually just regular number now to make it a bit
just regular number now to make it a bit
just regular number now to make it a bit more interesting we might have three
more interesting we might have three
more interesting we might have three nodes in a next layer here and then
nodes in a next layer here and then
nodes in a next layer here and then another three in the following one all
another three in the following one all
another three in the following one all of those being connected to each other
of those being connected to each other
of those being connected to each other and to keep it small maybe everything
and to keep it small maybe everything
and to keep it small maybe everything merges on the right-hand side back to a
merges on the right-hand side back to a
merges on the right-hand side back to a single one again now if you look at the
single one again now if you look at the
single one again now if you look at the from the from the larger picture here we
from the from the larger picture here we
from the from the larger picture here we have an input on the far left and then
have an input on the far left and then
have an input on the far left and then it goes to the right with an output and
it goes to the right with an output and
it goes to the right with an output and now let me first explain what happens in
now let me first explain what happens in
now let me first explain what happens in one of these nodes so um if we now look
one of these nodes so um if we now look
one of these nodes so um if we now look at anyone in their network in this
at anyone in their network in this
at anyone in their network in this particular one we actually have three
particular one we actually have three
particular one we actually have three inputs as you can see here and we
inputs as you can see here and we
inputs as you can see here and we typically multiply these inputs with
typically multiply these inputs with
typically multiply these inputs with some weights here shown by these W we
some weights here shown by these W we
some weights here shown by these W we sum them all up giving us a new number
sum them all up giving us a new number
sum them all up giving us a new number and that number we actually modify
and that number we actually modify
and that number we actually modify typically in a relatively simple way for
typically in a relatively simple way for
typically in a relatively simple way for example by cutting off the negative part
example by cutting off the negative part
example by cutting off the negative part and then the remaining positive one is
and then the remaining positive one is
and then the remaining positive one is sent on to the next node here in this
sent on to the next node here in this
sent on to the next node here in this case the output and that’s what’s
case the output and that’s what’s
case the output and that’s what’s happening in all of these nodes there
happening in all of these nodes there
happening in all of these nodes there might actually be many of them but
might actually be many of them but
might actually be many of them but there’s pretty much the same procedure
there’s pretty much the same procedure
there’s pretty much the same procedure everywhere now the trick here is that we
everywhere now the trick here is that we
everywhere now the trick here is that we actually collect a lot of in and output
actually collect a lot of in and output
actually collect a lot of in and output pairs and give that to a computer and a
pairs and give that to a computer and a
pairs and give that to a computer and a computer in a computation step can now
computer in a computation step can now
computer in a computation step can now figure out how to set all these weights
figure out how to set all these weights
figure out how to set all these weights so for each of the white arrows here we
so for each of the white arrows here we
so for each of the white arrows here we have one of these multipliers how to
have one of these multipliers how to
have one of these multipliers how to adjust some salt that we get the output
adjust some salt that we get the output
adjust some salt that we get the output as good as possible with that given
as good as possible with that given
as good as possible with that given network so that is the the actual
network so that is the the actual
network so that is the the actual learning step typically in these
learning step typically in these
learning step typically in these algorithms and yeah for that we actually
algorithms and yeah for that we actually
algorithms and yeah for that we actually need data again so if you think back
need data again so if you think back
need data again so if you think back about the example of Newton we might go
about the example of Newton we might go
about the example of Newton we might go back to observing falling apples
back to observing falling apples
back to observing falling apples but in this case it’s not enough to to
but in this case it’s not enough to to
but in this case it’s not enough to to see observe a single one but typically
see observe a single one but typically
see observe a single one but typically we need many of them so ideally we have
we need many of them so ideally we have
we need many of them so ideally we have millions of data points to use in this
millions of data points to use in this
millions of data points to use in this in this procedure and also thinking
in this procedure and also thinking
in this procedure and also thinking about this example of Newton what now
about this example of Newton what now
about this example of Newton what now happens here in a way is that this step
happens here in a way is that this step
happens here in a way is that this step have takes off the theory freaking out
have takes off the theory freaking out
have takes off the theory freaking out the theory and does that automatically
the theory and does that automatically
the theory and does that automatically in this optimization process so one one
in this optimization process so one one
in this optimization process so one one thing worth noting here is also that is
thing worth noting here is also that is
thing worth noting here is also that is not something that was recently invented
not something that was recently invented
not something that was recently invented by by one of the big Internet companies
by by one of the big Internet companies
by by one of the big Internet companies but it sounds in that if she comes from
but it sounds in that if she comes from
but it sounds in that if she comes from any centuries of a pretty fundamental
any centuries of a pretty fundamental
any centuries of a pretty fundamental academic research that now takes off in
academic research that now takes off in
academic research that now takes off in a way yeah so hearing about this in
a way yeah so hearing about this in
a way yeah so hearing about this in dnews you probably have read about the
dnews you probably have read about the
dnews you probably have read about the the superhuman performances and write
the superhuman performances and write
the superhuman performances and write one of this particularly good things
one of this particularly good things
one of this particularly good things here is finding cats and dogs and images
here is finding cats and dogs and images
here is finding cats and dogs and images so here are some examples where that
so here are some examples where that
so here are some examples where that would work very nicely and it’s
would work very nicely and it’s
would work very nicely and it’s wonderful in general because the
wonderful in general because the
wonderful in general because the Internet does probably consists of more
Internet does probably consists of more
Internet does probably consists of more than 90% of cat and dog images so it’s a
than 90% of cat and dog images so it’s a
than 90% of cat and dog images so it’s a great start em however often it’s also
great start em however often it’s also
great start em however often it’s also mentioned in conjunction with artificial
mentioned in conjunction with artificial
mentioned in conjunction with artificial intelligence and I have to admit I don’t
intelligence and I have to admit I don’t
intelligence and I have to admit I don’t like that label too much I mean it’s a
like that label too much I mean it’s a
like that label too much I mean it’s a somewhat philosophical question what is
somewhat philosophical question what is
somewhat philosophical question what is intelligence you can think about it as
intelligence you can think about it as
intelligence you can think about it as something taking an input and then
something taking an input and then
something taking an input and then producing some kind of reaction or
producing some kind of reaction or
producing some kind of reaction or output and but here also you quickly
output and but here also you quickly
output and but here also you quickly have these connotations of yeah machines
have these connotations of yeah machines
have these connotations of yeah machines turning against us Terminator style and
turning against us Terminator style and
turning against us Terminator style and when I think about all of these
when I think about all of these
when I think about all of these techniques is actually more about just
techniques is actually more about just
techniques is actually more about just representing a function really think
representing a function really think
representing a function really think back to high school f of X equal to Y in
back to high school f of X equal to Y in
back to high school f of X equal to Y in the end the computer figures out the F
the end the computer figures out the F
the end the computer figures out the F here which might sound very boring at
here which might sound very boring at
here which might sound very boring at first but that’s actually what makes it
first but that’s actually what makes it
first but that’s actually what makes it very genetic generic and also powerful
very genetic generic and also powerful
very genetic generic and also powerful because these deep learning algorithms
because these deep learning algorithms
because these deep learning algorithms are actually really better than anyone
are actually really better than anyone
are actually really better than anyone anything we had before to compute these
anything we had before to compute these
anything we had before to compute these functions and with that and now can
functions and with that and now can
functions and with that and now can explain a bit what I meant with the
explain a bit what I meant with the
explain a bit what I meant with the title so right going beyond finding cats
title so right going beyond finding cats
title so right going beyond finding cats and dogs and images I’m for example
and dogs and images I’m for example
and dogs and images I’m for example really interested in applying these
really interested in applying these
really interested in applying these techniques to these fluid problems and
techniques to these fluid problems and
techniques to these fluid problems and here now I can show a bit of
here now I can show a bit of
here now I can show a bit of research wave we’ve been doing in the
research wave we’ve been doing in the
research wave we’ve been doing in the past years on that topic and the first
past years on that topic and the first
past years on that topic and the first one is now focusing on smoke simulations
one is now focusing on smoke simulations
one is now focusing on smoke simulations so what we did here was and we looked at
so what we did here was and we looked at
so what we did here was and we looked at first kind of rough simple smoke
first kind of rough simple smoke
first kind of rough simple smoke simulation that you can see here on the
simulation that you can see here on the
simulation that you can see here on the left hand side with the simulation you
left hand side with the simulation you
left hand side with the simulation you can actually produce that pretty cheaply
can actually produce that pretty cheaply
can actually produce that pretty cheaply and then you can invest more time and
and then you can invest more time and
and then you can invest more time and produce a more accurate version of the
produce a more accurate version of the
produce a more accurate version of the same of the same smoke cloud here on the
same of the same smoke cloud here on the
same of the same smoke cloud here on the right we can see it has a lot more
right we can see it has a lot more
right we can see it has a lot more detail and structure now we looked at
detail and structure now we looked at
detail and structure now we looked at windows out of these data sets to make
windows out of these data sets to make
windows out of these data sets to make it a bit more generic and trained one of
it a bit more generic and trained one of
it a bit more generic and trained one of these deep learning use one of these
these deep learning use one of these
these deep learning use one of these deep learning algorithms to give us
deep learning algorithms to give us
deep learning algorithms to give us links or correspondences between these
links or correspondences between these
links or correspondences between these data points and then also once in the
data points and then also once in the
data points and then also once in the beginning we should go in and build a
beginning we should go in and build a
beginning we should go in and build a big library of these more accurate
big library of these more accurate
big library of these more accurate versions having done that what we can do
versions having done that what we can do
versions having done that what we can do is if we now look at a new smoke cloud
is if we now look at a new smoke cloud
is if we now look at a new smoke cloud for example this mushroom-shaped one
for example this mushroom-shaped one
for example this mushroom-shaped one that you can see here on the screen we
that you can see here on the screen we
that you can see here on the screen we can sample it with all these little
can sample it with all these little
can sample it with all these little windows load the library and now with
windows load the library and now with
windows load the library and now with the help of this strange network which
the help of this strange network which
the help of this strange network which now kind of takes the role of a smart
now kind of takes the role of a smart
now kind of takes the role of a smart librarian to tell us which data sets
librarian to tell us which data sets
librarian to tell us which data sets from the library are the the best ones
from the library are the the best ones
from the library are the the best ones that fit here and also respect the
that fit here and also respect the
that fit here and also respect the physics of this of this input we can
physics of this of this input we can
physics of this of this input we can create a version that right has a lot
create a version that right has a lot
create a version that right has a lot more detail and the date uses the data
more detail and the date uses the data
more detail and the date uses the data from the detailed and more accurate
from the detailed and more accurate
from the detailed and more accurate version here um I can directly show some
version here um I can directly show some
version here um I can directly show some examples in motion here this is now one
examples in motion here this is now one
examples in motion here this is now one of the inputs or we can see it’s
of the inputs or we can see it’s
of the inputs or we can see it’s relatively boring and the version you
relatively boring and the version you
relatively boring and the version you can see here now has this added detail
can see here now has this added detail
can see here now has this added detail from this from this library or this
from this from this library or this
from this from this library or this version shows some side by side the
version shows some side by side the
version shows some side by side the input ante and the more detailed version
input ante and the more detailed version
input ante and the more detailed version or another example that we simulated
or another example that we simulated
or another example that we simulated what is colliding rates of smoke here
what is colliding rates of smoke here
what is colliding rates of smoke here again this is the version with very
again this is the version with very
again this is the version with very little detail and the version with all
little detail and the version with all
little detail and the version with all the Edit complexity from the from the
the Edit complexity from the from the
the Edit complexity from the from the neural network here or side-by-side
neural network here or side-by-side
neural network here or side-by-side again even more obvious yeah so in a way
again even more obvious yeah so in a way
again even more obvious yeah so in a way this now gives us the more accurate
this now gives us the more accurate
this now gives us the more accurate behavior in a much cheaper in a much
behavior in a much cheaper in a much
behavior in a much cheaper in a much deeper way and also enables the the use
deeper way and also enables the the use
deeper way and also enables the the use of this library
of this library
of this library another example that we looked at were
another example that we looked at were
another example that we looked at were liquid simulations so here might
liquid simulations so here might
liquid simulations so here might actually look something like this a
actually look something like this a
actually look something like this a little blob of something like water
little blob of something like water
little blob of something like water falling to a little person where you
falling to a little person where you
falling to a little person where you cannot see the walls what we’re
cannot see the walls what we’re
cannot see the walls what we’re interested here in are actually kind of
interested here in are actually kind of
interested here in are actually kind of families of these simulations so hope
families of these simulations so hope
families of these simulations so hope you can see on the picture here that
you can see on the picture here that
you can see on the picture here that these drops change in size from small to
these drops change in size from small to
these drops change in size from small to large and also fall in at different
large and also fall in at different
large and also fall in at different positions and now if I show you the
positions and now if I show you the
positions and now if I show you the animation for it it gives you very
animation for it it gives you very
animation for it it gives you very different behavior so what’s here
different behavior so what’s here
different behavior so what’s here especially challenging is actually right
especially challenging is actually right
especially challenging is actually right how this behavior changes when we only
how this behavior changes when we only
how this behavior changes when we only change the size of this drop in the
change the size of this drop in the
change the size of this drop in the beginning and that is what we wanted to
beginning and that is what we wanted to
beginning and that is what we wanted to capture here so just to give you a very
capture here so just to give you a very
capture here so just to give you a very quick overview what we looked at we’re
quick overview what we looked at we’re
quick overview what we looked at we’re actually doing fairly expensive free
actually doing fairly expensive free
actually doing fairly expensive free computation step and then again training
computation step and then again training
computation step and then again training one of these deplane learning algorithms
one of these deplane learning algorithms
one of these deplane learning algorithms to if he kind of refined the behavior to
to if he kind of refined the behavior to
to if he kind of refined the behavior to take into account all the simulation
take into account all the simulation
take into account all the simulation data that we have in this whole space of
data that we have in this whole space of
data that we have in this whole space of simulations and what we get out in the
simulations and what we get out in the
simulations and what we get out in the end is actually a representation that is
end is actually a representation that is
end is actually a representation that is so small that we can run it on a mobile
so small that we can run it on a mobile
so small that we can run it on a mobile phone and you can actually try that
phone and you can actually try that
phone and you can actually try that yourself if you’re interested you can
yourself if you’re interested you can
yourself if you’re interested you can download the app that I’m showing here
download the app that I’m showing here
download the app that I’m showing here but this is a video of it running and
but this is a video of it running and
but this is a video of it running and yeah you have some additional pillar see
yeah you have some additional pillar see
yeah you have some additional pillar see around it but in the middle you can see
around it but in the middle you can see
around it but in the middle you can see these drops falling into the little pool
these drops falling into the little pool
these drops falling into the little pool and at different positions and sizes and
and at different positions and sizes and
and at different positions and sizes and if we actually try to run one of these
if we actually try to run one of these
if we actually try to run one of these simulations on the mobile phone it would
simulations on the mobile phone it would
simulations on the mobile phone it would probably take more than half an hour and
probably take more than half an hour and
probably take more than half an hour and he is running with 50 frames per second
he is running with 50 frames per second
he is running with 50 frames per second so effectively that’s probably a
so effectively that’s probably a
so effectively that’s probably a speed-up of more than 1,000 so it’s
speed-up of more than 1,000 so it’s
speed-up of more than 1,000 so it’s really a lot faster thanks to this
really a lot faster thanks to this
really a lot faster thanks to this thanks to the underlying algorithms and
thanks to the underlying algorithms and
thanks to the underlying algorithms and that kind of hints at at the
that kind of hints at at the
that kind of hints at at the possibilities here so we get a few very
possibilities here so we get a few very
possibilities here so we get a few very complex behavior very quickly and as far
complex behavior very quickly and as far
complex behavior very quickly and as far as we know it’s actually the first time
as we know it’s actually the first time
as we know it’s actually the first time you can actually run such a simulation
you can actually run such a simulation
you can actually run such a simulation on a on a mobile phone yeah and with
on a on a mobile phone yeah and with
on a on a mobile phone yeah and with these examples I want to take a bit of a
these examples I want to take a bit of a
these examples I want to take a bit of a step backward again and look at this
step backward again and look at this
step backward again and look at this bigger picture so I’ve talked about
bigger picture so I’ve talked about
bigger picture so I’ve talked about these different pillars of science and I
these different pillars of science and I
these different pillars of science and I now don’t want to claim that this deep
now don’t want to claim that this deep
now don’t want to claim that this deep learning is actually the fourth one or
learning is actually the fourth one or
learning is actually the fourth one or replaces the the computational side here
replaces the the computational side here
replaces the the computational side here but it definitely will be a very
but it definitely will be a very
but it definitely will be a very important component within those
important component within those
important component within those computations
computations
computations that’s pretty safe to say and right the
that’s pretty safe to say and right the
that’s pretty safe to say and right the key message really is that it’s
key message really is that it’s
key message really is that it’s something that is not just useful for
something that is not just useful for
something that is not just useful for finding cats and dogs and images but
finding cats and dogs and images but
finding cats and dogs and images but it’s a really generic computational
it’s a really generic computational
it’s a really generic computational method that influences the work of
method that influences the work of
method that influences the work of anyone working with computers and as
anyone working with computers and as
anyone working with computers and as such indirectly also all of our lives in
such indirectly also all of our lives in
such indirectly also all of our lives in the years to come so for my field of
the years to come so for my field of
the years to come so for my field of work what’s especially interesting is
work what’s especially interesting is
work what’s especially interesting is actually right capturing the behavior of
actually right capturing the behavior of
actually right capturing the behavior of nature around us forward so words
nature around us forward so words
nature around us forward so words ideally in in real time as realistically
ideally in in real time as realistically
ideally in in real time as realistically as possible and that starts of course
as possible and that starts of course
as possible and that starts of course was something like a falling Apple but
was something like a falling Apple but
was something like a falling Apple but ideally we’d like to look at effects
ideally we’d like to look at effects
ideally we’d like to look at effects such as the one shown here which is
such as the one shown here which is
such as the one shown here which is definitely an extreme example this is a
definitely an extreme example this is a
definitely an extreme example this is a giant wave hitting a lighthouse so if we
giant wave hitting a lighthouse so if we
giant wave hitting a lighthouse so if we could capture this kind of complexity in
could capture this kind of complexity in
could capture this kind of complexity in a virtual environment I think that would
a virtual environment I think that would
a virtual environment I think that would be really exciting and there’s potential
be really exciting and there’s potential
be really exciting and there’s potential here I believe and I don’t want to
here I believe and I don’t want to
here I believe and I don’t want to pretend that the skies are blue there
pretend that the skies are blue there
pretend that the skies are blue there are some definite clear limitations of
are some definite clear limitations of
are some definite clear limitations of these methods still today one of them
these methods still today one of them
these methods still today one of them for example is that these methods can
for example is that these methods can
for example is that these methods can actually not really learn larger
actually not really learn larger
actually not really learn larger contexts that well was kind of
contexts that well was kind of
contexts that well was kind of understand the bigger picture behind the
understand the bigger picture behind the
understand the bigger picture behind the data so that is something we definitely
data so that is something we definitely
data so that is something we definitely will need to work on and that we will
will need to work on and that we will
will need to work on and that we will need to improve in the future but
need to improve in the future but
need to improve in the future but whether we do that or not I think it’s
whether we do that or not I think it’s
whether we do that or not I think it’s already very powerful and ultimately of
already very powerful and ultimately of
already very powerful and ultimately of course the goal here is also to better
course the goal here is also to better
course the goal here is also to better understand the physics behind it not
understand the physics behind it not
understand the physics behind it not just display them but I think these
just display them but I think these
just display them but I think these methods can if you help us understand
methods can if you help us understand
methods can if you help us understand what is happening and one area where I
what is happening and one area where I
what is happening and one area where I think that is especially important that
think that is especially important that
think that is especially important that interesting is actually the the human
interesting is actually the the human
interesting is actually the the human body so if you think about the flow
body so if you think about the flow
body so if you think about the flow happening in your heart it’s actually an
happening in your heart it’s actually an
happening in your heart it’s actually an extremely complicated one and if we had
extremely complicated one and if we had
extremely complicated one and if we had methods there to quickly and accurately
methods there to quickly and accurately
methods there to quickly and accurately predict what the blood flow in a
predict what the blood flow in a
predict what the blood flow in a patient’s heart is that would be
patient’s heart is that would be
patient’s heart is that would be something that is really very useful for
something that is really very useful for
something that is really very useful for potentially very useful for medical
potentially very useful for medical
potentially very useful for medical doctors and could potentially save many
doctors and could potentially save many
doctors and could potentially save many lives so I think this is a particularly
lives so I think this is a particularly
lives so I think this is a particularly interesting direction for for future
interesting direction for for future
interesting direction for for future work yeah and I want to close with a
work yeah and I want to close with a
work yeah and I want to close with a quote from from Picasso and that I found
quote from from Picasso and that I found
quote from from Picasso and that I found he actually more than 50 years ago said
he actually more than 50 years ago said
he actually more than 50 years ago said that computers are useless because they
that computers are useless because they
that computers are useless because they can only give us answers in a way I
can only give us answers in a way I
can only give us answers in a way I think that’s still very true today so
think that’s still very true today so
think that’s still very true today so the interesting challenge really is how
the interesting challenge really is how
the interesting challenge really is how do we pose the right questions so in
do we pose the right questions so in
do we pose the right questions so in particular how can we actually step up
particular how can we actually step up
particular how can we actually step up questions such that these deep learning
questions such that these deep learning
questions such that these deep learning algorithms give us answers that we can
algorithms give us answers that we can
algorithms give us answers that we can use and work with and for me
use and work with and for me
use and work with and for me specifically I’m really interested in
specifically I’m really interested in
specifically I’m really interested in how can we set up these questions for
how can we set up these questions for
how can we set up these questions for the computer such that we’ll get answers
the computer such that we’ll get answers
the computer such that we’ll get answers about things like turbulence and
about things like turbulence and
about things like turbulence and understand it better and I think there’s
understand it better and I think there’s
understand it better and I think there’s actually a lot of very interesting work
actually a lot of very interesting work
actually a lot of very interesting work left so in a way exciting times for
left so in a way exciting times for
left so in a way exciting times for research and yeah with that I want to
research and yeah with that I want to
research and yeah with that I want to thank you for attention
thank you for attention
thank you for attention I also quickly want to mention this is
I also quickly want to mention this is
I also quickly want to mention this is of course based on work of my students
of course based on work of my students
of course based on work of my students so especially a Rachael and Lucas worked
so especially a Rachael and Lucas worked
so especially a Rachael and Lucas worked hard on the topics I showed earlier and
hard on the topics I showed earlier and
hard on the topics I showed earlier and if you’re interested in trying this and
if you’re interested in trying this and
if you’re interested in trying this and we have this app called neural liquid
we have this app called neural liquid
we have this app called neural liquid drop if you want to try these these
drop if you want to try these these
drop if you want to try these these droplets or real time liquids yourself
droplets or real time liquids yourself
droplets or real time liquids yourself feel free to try Thanks
feel free to try Thanks
feel free to try Thanks [Applause]
”
Be First to Comment