Heike Riel: How quantum computers will drive innovation | WIRED Smarter 2019

Good morning everyone.

Thank you very much for having me here.

Quantum computing, it's really

a very exciting new technology.

It's a completely new paradigm of computation.

It's using the laws of quantum physics.

So far we are used to classical computers

and we all use them.

You have them even in your pocket.

Quantum computers or classical computers

have really tremendously increased their performance

over the last 70 years.

We can do things today which have been unimaginable

a couple of years ago, and the fastest supercomputer today

is called Summit.

We actually delivered it last year

and it can do computations at a tremendous rate

of 200 quadrillion computations per second

or 200 petaflops, that's amazing.

However, there are still problems

which even the largest supercomputer cannot solve

and you may have some of them.

We can only approximate them

and we'll actually never be able

to build a computer which can solve those problems.

So a quantum computer is there to help.

This is a new type of technology.

We have invest a lot of research in the last 20, 30 years

to come as far and today they're here,

they exist in reality and we test them already

and they're available to you also through the cloud.

You see here the inner life of a quantum computer.

It's a very nice computer, it's a beautiful picture

and the quantum processor is actually housed here

at the lower part in the cylinder.

There are lines coming in from the outside world

to go to the quantum processor.

These are superconducting coax cables

where microwaves are used

to actually control the quantum processing to bring in

and bring out the information.

All this is housed then in a dilution refrigerator

to cool down the quantum processor to very low temperature

to about 15 millikelvin

or 100 times colder than outer space.

Because the quantum information in these qubits

is very fragile and delicate

and we have to avoid high temperature

or also light electromagnetic waves,

which could disturb the quantum computation.

So we have to keep it very cold and very dark.

This is the heart that's a quantum processor

and these chips are built and designed within IBM.

This is an example here of a 16 qubit processor

where you have these square there, these are the qubits

and these wavy lines are the microwave resonators

to talk to the qubits and control the quantum calculation.

This is not a transistor,

it's not based on the classical laws,

but really the laws of quantum physics

and therefore it can do amazing thing.

It's really a game changer, which is coming today.

And why is it a game changer?

You all know classical bits.

A classical bit can be a zero or a one.

It can be also represented by an arrow up or an arrow down

and all the information is encoded

in strings of zeros and ones

and then we can do calculations with them.

In a quantum bit, however, it's different

because the quantum bit can be a zero,

a one and both at the very same time.

So a quantum bit can actually represent all the information

or all the different points on the surface of this sphere,

which I illustrated here

because quantum mechanics uses the laws of superposition

where we can use these up vector and down vector

to represent the information.

So it's a much richer information space

we can use for calculation.

So this makes a difference

because it gives us exponential power for quantum computing.

What does it mean?

If we have one qubit, we have two states,

and this is the superposition of these two states,

which we represent as a zero and a one

and alpha and beta are factors.

If we have two qubits,

then we have actually four basis states

because we have the zero, zero, zero, one, zero

and one, one.

If it takes three, you can imagine what happens.

Now we have eight basis states.

So if we go up and go to a 50 qubit system

which exists today,

then you have actually one extra basis state

or one million giga basis states,

which is already a very large number.

So you see that the dimension which we can then use

for calculation is exponentially increasing.

If you go even further up to 275 qubits,

you have more basis states

than there are attempts in the observable universe

and this is actually why you can never build

a classic computer with this number of information

you can encode in.

So this is exponential scaling

and a quantum computer exponentially scales a performance

by two power to the N where N is the number of qubits.

So if you compare this to a classical computer

as shown here, then in a classical computer

we have today, this is an example of the IBM power nine,

which about eight billion of transistors.

If you add one transistor more, it doesn't matter too much.

If you want to double the performance,

you need to double the number of transistors.

However, for a quantum processor, it's different

because you just need to add one further qubit

and you can double the performance in the ideal case.

So this is really the potential of quantum computation,

which we are all after.

So what do we wanna do with it or what can you do with it?

So there are many problems, mathematical problems,

which exponentially increases in complexity

and these are the problems where quantum computers

are built for,

for difficult problems in business and science.

We know the easy problems.

We can do a multiplication with small numbers in our head.

For larger numbers we may use a classical computer,

but there is an example like factorization,

a small number like 91,

we may be able to do this also in our head,

but if you go to very large numbers

and want to factorized them, then it's very difficult.

And this keeps also very difficult for a classical computer

because it has to go through one by one.

However, for a quantum computer, that's an easy case.

And these are the types of problems

which quantum computers can have an advantage in the future.

And so there are problems in materials and chemistry

where you implement or encode your molecular state

into a quantum computer,

also in machine learning and in optimization.

So let's have look at a few of those applications

and let's start with classification.

In classification, you all use in the finance industry

because machine learning is used in every industry today,

it has a lot of impact

and it's already used for applications

like fraud detection, for credit risk rating,

or also customer segmentation or anti-money laundering

and more examples.

And this is an example.

Let's say we want to decide

whether we wanna give a customer a credit or not.

So you want to separate the credit worthy

and the credit customers and you want to separate them.

So in this case, which is illustrated here,

it can be very nicely done.

You have both of these groups, the orange ones,

the credit risk and the ones you wanna give a credit to,

you can very nicely separate.

You have a nice hyperplane in between

and you actually want to maximize the margin

between both groups that you don't make false positives

or false negatives.

But what if the data is not linearly separable

as in this case?

So let's assume you have this upper line

where all these dots are mixed with each other

and you cannot find a nice line to separate both groups.

So the trick you do to tail already

is you go to a higher dimension as shown in the lower part

of the picture where you go from the linear

into the space, into the area,

and then you can nicely separate

with both groups with each other.

That's a very simple problem.

However, there are more problems where this is not as easy

to be done and actually quantum mechanics can help

because you have this complex room

which quantum bits can offer you.

So you can lift your problem into a higher dimension

where you can use quantum feature maps

to separate those groups.

And we have tested this on a small quantum computer

in this case with its supervised learning

with quantum enhanced feature spaces

and we have demonstrated this example

on a real quantum computer for small test cases

where we demonstrated a potential advantage

because we could increase the accuracy of the segmentation

of the classification and we actually could also prove

that the classification accuracy improved

with increased entanglement of the qubits.

So this is one example where quantum computing

can provide a true advantage.

Another example I want to give you

in the pricing and risk analysis

because these are also very complex problems

where you spend a lot of computation

and cost in order to do pricing and risk analysis.

So in this example,

you are interested of course in the value at risk

and also into the conditional value at risk

of your portfolio.

And you typically do Monte Carlo simulations

in order to calculate this risk.

And these Monte Carlo simulations,

depending on the complexity of the problem,

may take you overnight to calculate a decent accuracy

of your portfolio or they may even take longer.

So we have actually developed now a quantum algorithm

where we can speed up this type of risk analysis

by quadratic fashion.

This means if you need to draw one million of samples

in your Monte Carlo simulations,

then you only need to do about 10,000 calculations

for a quantum computer so you get a quadratic speed up.

And we also demonstrated this on a hardware,

on a small hardware with a small number of quantum qubits.

And you see here the estimation error

where we prove that the estimation error goes

faster down than in the Monte Carlo simulation

and already with a small number of samples,

you have an advantage.

So of course these are at that moment right now,

small samples and small problems

where we apply quantum computers.

But this has to do also with the size

of the quantum computers currently.

Another example where also Monte Carlo simulations are used

is in the area of option pricing.

And also here we actually have worked together with JPMC

to test the algorithm,

which we have developed for option pricing.

And in this example,

we have used European call options with drag point K

and you see here in red the payoff function

and the spot price distribution in green.

And we have actually implemented this example

on a quantum computer and varied the spot price

and also have seen the quadratic speed up

of the calculation.

So let's move on and to look at combinatorial optimization,

that's another very important problem

where quantum computers may be able to help in the future.

And in this case, we have worked together with Barclays.

And the example which has been used here

is a quantum algorithm for mixed binary optimization,

which was applied to transaction settlement.

And so clearing house has a very complex task

because they have to clear continuous incoming transactions.

And the better they can do this, the more value is created,

the lower the risk is.

And so in this case, we have studied an example

also with three parties.

And you see here the different transactions

which are possible, T1, T2, and T3.

And you see at the bar chart here,

the calculation of the quantum computer,

it was done on a five qubit quantum computer

and it clearly shows that the probability

is the highest for the best solution,

which says that we want to do the T2 and T3 transaction

and cannot do T1.

So we are right now extending these algorithms

to make binary optimization problems,

which enables new applications.

So let's move on and show you

where we are in quantum computing.

This is a system which we have introduced

at the beginning of this year.

It's the first IBM Q System One,

and it's really now moving out the quantum computers

from the lab environment where we as physicists

optimize these devices to a environment

where we put them in data centers where they're optimized

for stability, for reliability

and continuous commercial use.

And so just recently on the 18th of September,

we have actually brought 10 quantum computers

into a data center and they're open via the cloud.

And so we are building also an IBM Q user community

and also a network where we work together

with Fortune 500 companies, universities,

and other partners

in order to further advance quantum computing,

to educate and also to launch

the first commercial applications.

So that's a very exciting time right now.

There's a lot of activity and things happening

and we see more applications coming up.

And with this, I would like to stop

and I would like to encourage you to also explore the world

of quantum computing

and you find more information also online.

Thank you very much.