Dan: Hello and welcome
to the quantum divide.

This is the podcast that talks
about the literal divide between

classical IT and quantum technology,
and the fact that these two domains

need to become closer together.

Quantum networking actually is
more futuristic than perhaps

the computing element of it.

But we're going to try
and focus on that domain.

But we're bound to experience many
different tangents both in podcast

topics and conversation as we go on.

Enjoy.

All right.

Welcome back.

Another episode in short succession.

This time I'm joined by
Harold Olivier, right?

Bonjour Harold et bienvenue
sur le podcast Quantum de Vide.

Thanks for joining us.

Yeah, superb.

Yeah.

Yeah.

You're you're a first
attendee from France.

So it's great to, to have you on
board and we'll have a good discussion

about what you do and what's happening
in France, I think is superb.

So you work for Inria, right?

Um, Yeah.

Um, Let's start with a normal approach
that I take in these podcasts, give

us a bit of a view on your path
into quantum what it is you do and

what it is you're doing at Inria.

Harold: So what I'm doing is
mostly researching um, verification

of quantum computation.

That means when you are in a setting
where you don't have a quantum computer

in house and you want to delegate
the computation but you want to.

this computation to still be secure
in the sense that you want to

have privacy on the data and the
algorithm that you are running.

And you also want to have
a proof of integrity of the

computation being performed.

So that's most of my activity,
although I have other interests

um, mostly based on quantum error
correction, quantum error mitigation.

And the reason is that the team that
uh, I'm working in is uh, its focus is

really working and researching theoretical
problems fundamental problems, but guided

and inspired by practical problems,
experimental questions and trying

to make that real life applications.

So we go back and forth
between theory and experiments.

So of course, if you do this,
you are forced to look at error

correction and error mitigation.

So that's, yeah, that's the second

Dan: focus.

Great.

Is your background in mathematics or is
it physics, a blend of multiple topics?

Harold: Yeah, it's mostly a blend.

Yeah I had, a lot of math, but but
also I did, I started doing theoretical

physics and I entered into quantum.

Really through the theory of decoherence.

So really, like I was interested at
the time in studying the transition

between quantum world into the
classical world and what that means.

And in fact, I went to this topic or
I entered quantum information because

The work that I did was trying to
understand this transition mostly from

an information theoretical point of view.

So really trying to understand
how information leaking into the

environment leads to decoherence.

And more than this is like, how does
objective world emerge from the quantum

one and again, trying to characterize
this in terms of information.

And yeah, it's very natural to then
input a lot of quantum information,

theoretical quantities into that.

And then like that, yeah, you start
being interested in error correction

and then goes to more like applications.

Dan: Great.

I know we're going to deep dive
on some of that in a moment.

But first of all if I can just take you
back to Inria, so I understand you're a

leader lead for the quantum activities
within Inria research organization.

It seems that Inria focuses on
many different industry verticals.

Is there a relationship with
the French government now or the

French national quantum strategy?

Harold: Thank you.

Okay.

So Inria is one of the various
research institute, public

research institute in France.

The specificity of this institute
compared to what usually people

know more abroad is CNRS.

CNRS has a very wide and broad interest in
science they, they can go from archaeology

to to literature, to history, to also like
computer science, math and everything.

INRIA is focused only on computer
science so within this focus,

INRIA has nine different centers
in in France and many more centers.

That are in partnership with universities
and they're in these in these centers,

you have like smaller teams it's
usually between three and five, maybe

up to eight people that work on a
topic and it's it's yeah, each team

needs to have a rather clear topic,
clear focus for the next, Decade.

And yeah, we work then on
various topics within Inria.

So you have about 1500
researchers within Inria.

There are several teams doing
quantum information science.

Right now it's about 80 researchers
and on these topics we look

very much into, I would say the
core of the quantum ecosystem.

So we think about.

We have teams doing pure
quantum information science.

Some doing languages compilation.

We have error correction.

But we also have partnerships and
that's the strength of the model.

We have partnerships with
universities that created teams

that have both experimental
physicists and computer scientists.

So one of the team is at the origin
of the Cat Qubit model for quantum

computing, that has gained a lot
of interest in the recent years.

And so it's very interesting because
It's within a computer science

institute, but you still have
like dilution fridges in the lab.

So that's very interesting in there.

So that, yeah, our focus is a lot onto
what can we do with quantum computers?

How do we build these quantum computers?

How do we make them robust and trusted?

So that's, that, that's I would
say the main focus of INRI.

Dan: Very nice.

Thank you.

Yeah.

How does that link into the
French quantum strategy?

Is there a national strategy and
a good ecosystem of startups and

things that you collaborate with?

Or is it heavily public sector

Harold: driven?

Yeah, so I would say so the topic
the topic of quantum information

at Inria started in 2001.

Actually, I was I was the first one
doing quantum information there.

And it grew a little by little
and, yeah, as a tradition or uh,

yeah, recognize expertise in, in
having collaborative projects with

with startups and with companies.

Since the very beginning, there has always
been a lot of contact with industries

in quantum, of course, and others,
but not necessarily the quantum topic.

But yeah, there has always
been a lot of interaction.

Also, because these teams are often within
universities, there are natural links,

and I would say it's the it's a very
strong characteristics of the of the field

of quantum information science is that
theory people usually talk a lot more with

experimentalists than in other fields.

And so there has always been a lot of
interaction with with the other academics

team in the ecosystem, but definitely
with the launch of the national quantum

strategy, like there has been a lot more
of effort done in this in this respect.

Coming from the state and the government.

And INRIA is one of the three main
research organization in France.

Leading and orienting the national
strategy with CNRS and CER, and

yeah, we actually, before joining
the podcast, I was in a meeting with

them and we were discussing like.

How to put a lot of synergy within
the different projects that are

funded at INRIA, but within the
national strategy, but also beyond

we are we are all scientists working
in this way of orienting things

and We fully appreciate the effort
that has been and the trust that is

put on us by the government saying,
okay, develop something that works.

That's a big challenge.

But at the same time, we know that by
construction, these kinds of national

strategies have a very short horizon
even though for them, it's a huge

effort going to, looking at what's
going to happen in the next 10 years.

But in terms of research, it's still
like short, medium term, and we

need to plan what goes ahead of this
and what goes further than this.

And our discussion is okay.

We want to take care of this opportunity
to, yeah, go on and be on target with the

challenges that are assigned to us by the
by, by this strategy, but at the same time

construct something that goes beyond this.

And that's, yeah, a lot of our
discussions with our colleagues.

And of course.

There is.

How do we create something that's really
like very strong academic ecosystem.

But as well, like, how do we
partner with companies and startups?

There are several startups in France that
have been quite successful at raising

funds in building quantum computers.

We start having startups also
interested in building quantum memories.

And we go then more towards
networking questions.

There are people doing solutions
for for right now, like distributed

classical computing, but secured
with the with quantum links.

So yeah it's growing in several
directions and it's it's very

active, so yeah, very encouraging.

But we are always interested in also.

Opening this ecosystem to welcome
people also from abroad, because

we think that it's necessary France
is not big enough as a country to

produce everything that's needed.

So we need to be in contact with our
European partners, but also more wider

international partners on that topic.

Dan: Yeah, exactly.

It's definitely a global
global effort at the moment.

Although there's still the whole
local political things at play, but

I think the scientists try and do
their best to ignore that, right?

And focus on the science,
focus on the technology.

But I'm not familiar with some of
the startups and companies in France.

Do you want to list a couple?

Harold: Yeah, I hope
I don't forget anyone.

In quantum quantum computing, there
is one company, Pasqal, which has

been quite successful in developing
a platform for neutral atoms.

That's one.

There is one that's actually
very much linked to with Inria.

because of this team developing
CatQubits, which is called Alice and Bob.

So that's the superconducting
platform for computing, but

really based on these CatQubits.

And they aim directly at
going to fault tolerance.

So that's that's interesting because
it's a slightly different roadmap.

And the way they develop things is of
course different than the others because

they are not so much interested in NISQ.

We have Candela which is a very
interesting company working on

photonic quantum computing platforms.

There are younger ones C12, which is
working on carbon nanotubes for doing

the the shielding, I would say of
electron spins for quantum computing.

That, that's I think pretty
much what we have more, most

advanced in the computing space.

We have Willink, which is a startup
that has started the working on

quantum memories and VeriQloud,
which is this this company securing

classical computations with quantum
before being in a position to secure.

Quantum computations as well.

So I think that's the ones that I'm
thinking about in terms of really hardware

computing or networking hardware, but
we have other companies doing cryostat

and electronics but that's a little bit
too far from me to be really listing

them because because yeah, we are still
working at a quite theoretical level.

So okay.

Everything that goes around, which
is extremely important, is slightly

out of my scope, I would say.

And yeah.

Dan: Totally understand, yeah.

No, thanks.

That's interesting that even though
you mentioned most of those are

hardware orientated, I think there's
a software platform there's...

Obviously to create a fault
tolerant system a lot of that

is software orientated as well.

Harold: Yeah, that's true.

There are a few startups doing software.

And now I'm missing the names, of course.

But yeah it's also fairly active.

I would say by construction,
it's the these startups.

need less capital and it's easier
to develop this kind of of solution.

It's a little bit less in the focus maybe
of of the national strategy, although it's

very important in the long run, because
That's the way you actually reach the end

user and really know who is needing what
and what quantum computers are useful for.

So that's very important
and they are very necessary.

Also like for us as a
research institute because.

We are sometimes contacted by large
companies who want to have or understand

quantum, and it's hard for them to
find someone who is going to really

that progress in understanding what's
at stake for their organization.

And a lot, most of the time
they come to us and they ask

for help and we try our best.

But at the same time,
it's very hard because.

As a research organization,
we do research.

And so everything that we do So

you need to do science at the level
that's really like state of the art

and really pushing the boundaries.

And sometimes there is a misfit between
between what we should do and what the

companies can, swallow and understand.

So their companies doing software
are very important because they are

good intermediate level, like to help
prototyping some stuff and help these

larger companies grow in expertise.

And then it's much easier to do like very
advanced projects with them because they

already know the basics and they know what
they want on or where they want to go.

And it's much, much easier
as far as to interact.

Although we try to do our best.

With addressing these issues
unfortunately, we have a

limited bandwidth in some sense.

We need to deliver the research that
we promised to the funding agencies

and the various contracts we are in.

So we do our best in the free time
or allowed allowed times there.

But but when we do it, it's very
interesting because it's It's really

opening different perspectives
and and making that more connected

to sometimes like real use cases.

Sometimes it's also achieving
very unexpected results, like

for instance, helping a company
improve its classical algorithmics.

Sometimes they come with a problem and you
find a solution that doesn't need quantum.

And that happened to me and I
think it's very interesting.

Because I think this is an
effort that everyone needs to be

starting performing right now.

Because of the end of the Moore's law,
we need to go back to algorithmics

and try to fine tune everything.

And that's, that's a very hard
challenge, but that goes way beyond

quantum, but if quantum can be one
of the reasons why you want to open

the box and look at the things like
reading in detail you should do it.

And if you find a good classical
solution, it's all the best.

Dan: Yeah.

I say that's a surprise, but a
nice surprise, I guess, um, it's,

I feel like we could talk about the
business and ecosystem for the whole.

call.

But I know we came to with
three main topics, right?

We wanted to talk about the trends
in quantum computing, HPC and data

centers and that kind of stuff.

We wanted to talk about privacy,
security, and trust in quantum

computing and where that was necessary.

And then of course, quantum networking,
our favorite topic at the end.

So let's dive into that, shall we?

Let's start with the first one.

We had a conversation about quantum
computing on premise consumption of cloud

deployments, some companies looking for in
house systems and then also the challenges

that brings for a data center company.

Yeah.

How would you like to start on that topic?

I know that's an interesting one for

Harold: you.

Yeah, for me, it's really understanding.

Okay, there are a lot of challenges
in developing quantum technologies.

There is of course the science of it but
it's not only this At one point we will

be developing this and people are going
to sell stuff to others to companies

and to sell stuff I Mean I have been
a venture capitalist in a past life.

So I know that if you have a very good
product that's fine but you need to

meet the market and you need to really
answer or, yeah, you need to answer a

need and of of a user and you need to
insert yourself in the way he processes

things and in the way he deals with,
you This service or this hardware

that you're providing, and that's,
that's, that's a more complex or it's

a wider concern that you need to have.

And what I would say is that
right now in quantum computing, we

have always been referring to the
early days of classical computing.

And that's partly true, but it's also.

Sometimes a little bit misleading.

Of course our machines are not efficient,
not very powerful and they are super

big and yes, there is a long way to go
and we still need to work and okay sure

but at the same time there is something
that has changed between the time of

ENIAC and now is that now it's just For
most companies, not an option to have

in house computing like super very heavy
computers and things in house because

we are now so much accustomed to sending
data around and moving it here and there

and then, okay let's hire computing
power for a peak demand at one point.

And then we release that.

This is common practice
in large organizations.

Okay.

And if we arrive with a very nice
and good computer, quantum computer

now will be a pain for these people
because they need to reverse what they

have been doing, which is like pushing
the hardware outside of the buildings

and into specialised data centres.

And we will, if we are selling
them a product that needs.

For them to reverse this trend,
it's just like almost impossible.

So I think we need to
take that into account.

And people providing services as data
centers need to take that into account

because they need to, at one point, be in
a position of hosting quantum computers.

And we see this in some of the projects

that we are that we are
developing and deploying in

France for the national strategy.

. Which is we are trying to host in one
of the public data centers some early

quantum machines, and it's complicated.

It's complicated because the
requirements of these machines are

very different from the classical ones.

Especially like the cooling system
with helium is a pain and data

centers, they don't want this.

So they, they actually need to rebuild
rooms for us to put the machine.

So it's like suddenly you are like very
much on to brick and mortar questions.

But you need to solve this.

So that's one, one problem.

The cooling, there is also the vibration.

In fact, data centers are vibrate a lot
and quantum computers don't like that.

So you need to take care of this.

And again it's a problem.

Hiring construction companies, redesigning
your building and things like this.

So it's a pain and it's very costly.

So I think it's good that.

We do it like with public money
to see exactly what it is, what it

takes, understand the problems and
then transfer that, to more like

to the private sector at one point
and say, Okay, here is what we did.

This works.

This doesn't work.

And we can try to help you
in addressing these issues.

Yeah.

So that's that's what they would
say regarding data centers.

I think it's, we need to take
into account the fact that people

don't want machines in premise.

We need to adapt data centers
to host quantum machines.

And then yeah, of course, then comes
the question of what security for this.

Dan: That's a different matter for sure.

Yeah.

So at the moment when a quantum computer
is installed somewhere, do these

types of physical aspects need to be
considered and implemented every time?

Maybe it depends on the modality, right?

So if you need a dilution fridge,
then of course you have the

chemicals required with that.

But I do hear that these days
the dilution fridge technology

is pretty more off the shelf.

Type type products.

Is that true or

Harold: not?

Right now, what we what we see
is that, okay, I would say that.

We are conducting experiments.

So if we do something, we take the most
demanding platform and we do something

that fits this platform and then we'll
host in the same space the other machine.

Of course, there are technologies that
are less demanding than the than maybe the

ones you are requiring dilution fridges.

But still, it's something that needs to
be right now it's still, it's apparently

it's still needs to be built, pretty much
on purpose for this kind of environment.

But it's also, it's also a question,
like very practical question of these

machines are experimental machines still.

So we think we know that we will want
to open the box from times to times.

So we want to have
access to these machines.

We want to be able to
work around the machines.

We want to be able to have enough
space to in case we want to upgrade

some components and things like this.

So, so, Yeah, I'm not sure
we can really completely draw

conclusions from what we have.

We are seeing right now because
we are in an experimental phase.

And but in a few years from now, we'll
have the, the probably some much better

view or a clear view on what is needed,
what is not needed, but we did it

anyways by abundance of of precaution
but yeah, so I think it's still

there are still like open questions.

We just want to make sure.

That we can push these experiments
as far as we as we can.

We don't want to be blocked at one point
by something that we didn't think about.

So we are a bit too cautious probably but
yeah, I think it's fine in this stage.

Dan: And that leads me to think that
there's a link between this topic and the

second topic about privacy and security.

Because people that want to use
quantum computing resources.

And don't have the ability to either
build a new building, or take on some

new chemical processes, or hire some
physicists to look after the computer.

They're going to need to think about
some form of remote execution of

algorithms or functions inside their code.

We know there's already many ways to do
this through the private quantum computing

companies, through the public clouds.

Yeah.

This is a this kind of part of the
industry has grown a lot quicker than I

think the actual quantum technology itself
in the back end because of the fact that

the cloud networks are everywhere, that
the cloud data centers are everywhere.

It's a matter of, pushing out
new functions and they get

distributed around the globe.

That's great.

But then and we enter the topic of what?

What if you're executing something
that has some intellectual property?

What if you're testing something that
is you want to keep secret or you don't

want anybody to know the algorithms
you're running and testing to maybe

solve a particular new problem that
leads us on to the next topic, right?

Harold: Yeah, I think it's I think
it's a trend that is actually coming

from outside quantum computing really
with the privacy preserving laws,

especially in Europe with GDPR everybody
now knows that there is a risk in

mishandling confidential private data.

People are now also very much aware
of Intellectual property rights and

questions that you have when you
actually execute things remotely.

And right now, even in the classical
crypto field, there is a lot of evolution.

The mentalities are changing, and
people are now much more aware of the

issues and the potential solutions.

And in the classical space, there
are a lot of solutions but it's

still very early, I would say.

There are plenty of solutions,
and people are still exploring.

For instance, not so long
ago, we were discussing with

the French banking regulator.

And they were exploring.

Like very basic questions like, okay,
we want to conduct stress tests on the

banking system or maybe the insurance
insurance companies, the way they

conduct stress test is that they need
to have a very fine grained information

about all the contracts and all the the
commitments the banks or the insurance

companies have with respect to the various
parties they are doing business with.

The problem like if you take like
for instance an insurance company

the commitments that you have is
Towards actual physical people, okay?

And to understand how this contract
is going to behave, you'll have to

transfer a lot of private information
about the health of the person, about

whatever, the age and location, wealth.

So it's a nightmare, even for regulators.

That's really a, a, a, a
difficult topic for them.

And they are exploring
classical solutions for this.

The way they do is actually hire
various startups and make them

work on these kind of issues.

And so we contacted them and saying have
you considered quantum computations?

Oh, yeah that's another challenge.

We are interested in this
but it's not for now.

And we all agree that they are not going
to conduct this kind of test with quantum

uh, quantum computers anytime soon.

But then we, we asked them
like, but are you going to

have your own quantum computer?

Oh, no, of course not.

We are a regulator.

We will just, do that on the cloud and say
but then you just have the same problem

as you have in the classical space.

And then they just realized that.

If you think about a lot of quantum
applications you'll be faced with the

same questions that you have right
now in the classical computing space.

And so we need, as scientists,
we need to develop these kind of

solutions or protocols to make
sure that we can guarantee the same

kind of of promises to end users.

And I think for me, it's really
like Uh, very strong requirement for

developing a healthy quantum ecosystem.

If we don't have this Then, and I'm
coming back to what I was saying

before, if we don't have this, we'll
have superb machines, they are going

to be great, but nobody or a very
small fraction of what we have in mind

as a market will really be able to
do the efforts to be using it because

you need to have a new building and.

You are just renting your space or
maybe you cannot build something

inside to host a quantum computer.

You can have like plenty
of reasons like this.

So I think it's necessary if
we want to unlock the full

potential, we need to make this.

Remote quantum computation available
if we do remote quantum computation,

because quantum computation at least
for the first, first wave of things

that we will be doing, it's you
are going to compute with a quantum

computer, the most complicated or the
most sensitive data that you have on.

So by construction, those will be
Requiring a lot more protection

than a regular computation.

So we need to address that question.

Yeah, that's

Dan: for me.

Yeah.

Yeah.

I need some help.

I think so.

My just in understanding part of that.

I understand the privacy and the
need to execute in a secure way.

But my understanding of quantum
algorithms mostly comes from Grover's

search and Shor's algorithm, so on,
which are executing a particular.

set of mathematical tasks to
find an answer to something.

None of that really takes, of course,
if you're factoring a number and it's

a key, which is to something important,
then that's those numbers that you end up

with are quite important and may be seen
as something you need to keep secure.

But what about, you mentioned
the insurance company, you

mentioned personal information.

Even more so if it's
company private information.

Can you give me an example of how
that type of information is used in

quantum algorithms or programs which
are executed on quantum systems?

Maybe we're not there
yet, but I haven't seen...

Harold: All the people that are
doing quantum machine learning,

they are learning on something.

And so this something might
be, whatever you want.

But you cannot rule out the fact
that these data are sensitive.

That's one possibility.

The second thing is maybe it's the
algorithm that is sensitive itself.

Okay, so you have found a nice way
to do something and you don't want

your competitor to be using it.

And my most beloved
example is Always the same.

And I'm sorry for naming companies, but
maybe it's just making things clear for

people listening to the to the audio.

But if I am Criteo, which is
like an ad placement company.

And that I have found a quantum algorithm
to optimize this ad placement, and

I'm running this against Google, okay,
because Google is my competitor, or maybe

Google is running the platform where I'm
trying to find the best way to place ad.

So I have a competing interest
in this ad market with Google.

But will I really delegate my
quantum computation to Google

cloud quantum cloud services?

I don't know.

But Definitely, even though the
data is not sensitive, I don't

want to give them my algorithm.

Okay.

And the way it could be reversed,
if Criteo had developed quantum

computers, and maybe Google
doesn't want to use that for him.

Okay.

So So I think that's that's the issue.

The algorithm itself
can be very sensitive.

And right now, when you access
when you access a quantum computer,

you just send your instructions
in plain to the quantum computer.

So if you are sitting on the other side
of the quantum computer and just looking

at what instructions are executed, you
can just Find out the whole computation

that is being executed and we know this we
experience this Because we use of course

it's in an agreed way with the various
Computing Providers so as academics we

often have access to their machines, but
we agree this to be open And we know that

they look at it because from time to time
we get a call or an email saying, Oh, you

are running something and we quite don't
understand what you are trying to do.

So it looks interesting or something.

So it's a fact.

You can really look at what people
are executing on quantum machines.

And I think it's good.

I like it the way they do it.

I'm not complaining about this.

I need to be clear on this.

I think it's good because we need
and we want interactions with the,

the hardware vendors and to, to.

Help us achieve what we want to
do on their platform, but at the

same time trying to help them
develop better and better tools.

But so that's a well agreed situation.

But at the same time, if you are in a
different situation where you want to

have your confidentiality well, You
need to rely on the fact that the people

on the other side are are uh, okay and
uh, are following their obligations.

And that's, most of the time, that's what
you don't want to do when you think about

Dan: cryptography.

Yeah, of course.

So there are pros and cons of
the, of the, basically, feature.

I mean, In order to execute something
on a remote quantum computer, you need

to either describe the entire processing
gate form, or you need to Describe all

of the mathematical operations that
you need to make on the qubits, right?

Yeah.

Yeah.

Okay.

What options are there to
obfuscate, those instructions?

Is there such

Harold: a thing?

Yeah.

That's, now we are getting really
like onto what what interests me as

a researcher is really like this is.

This ability that we have to
hide the computation and the data

from from the computer or from
the server that is executing it.

And if I want to put it very simply
in simple terms basically the way it

works, there are several ways to do
it, but one way of doing it, which is

easy to understand is to say that if
we want to conduct a quantum operation,

or if I want to instruct a server
to do a quantum operation on a qubit

that I send to the quantum computer,

I need in this case to agree with the
server on a common reference frame.

Okay, so it's basically an
orientation of the axis.

Okay, so if I want to even think
about quantum cryptography,

which is even maybe simpler.

If you execute the BB84 protocol,
you need to send qubits polarised

vertically, horizontally, or
on a diagonal basis, okay?

So if I do this, and I send, as one
of the parties, I send my qubit, and

I claim that this qubit is polarised
vertically I need to agree that

my vertical is the same vertical
as the one on the receiving party.

And everybody knows that it's a highly
non trivial fact, because if you put,

especially if you put your qubit in
a fiber, it's going to rotate, okay?

So you need first to actually um,
uh, calibrate your orientation of

axis between one party and the other.

If we have some quantum data and we send
it to a server to process it, We will

need to share a common reference frame.

Now, one way of hiding the
computation, one way of hiding

the data is quite simple.

What I'm going to do is that I'm
going to choose one reference frame

for myself, which will be secret.

Okay.

And then I'm going to actually
synchronize a relative reference frame.

With the server relative to
my own secret reference frame.

Okay, so what's going to happen is
that I'm going to send instruction to

the server in the relative reference
frame, it's going to correspond to

the operation that I want to perform
on my side on my secret reference

frame, but the server is just going
to see very random instructions,

because actually what I'm going to
do is that I'm going to change this

relative reference frame qubit by qubit.

Okay, so I'm sending, I'm agreeing with
the server for a reference frame, but

I'm keeping constantly changing it and
he doesn't know how I'm changing it.

Okay, and that's the way I'm hiding
everything, all the computation, all

the instructions and also the data.

So it's very simple.

In terms of physics it's much,
much more demanding if you really

want to build it as a hardware.

But the principle is simple.

And then from there, you can
actually do a lot of proofs and very

interesting proof because it will
give you security in an information

theoretical sense, which means like
really The best that you can hope for.

So that's one way of doing it.

There are other ways of doing it.

People are thinking about trying to
build things that would be similar

to enclaves in classical processes.

They would call it quantum enclaves.

Okay.

So to try to avoid sending qubits from
the client to the server, there are

people that say, okay, let's rely on
post quantum security uh, cryptography,

sorry and uh, build this concept
of enclave but on chip uh, through

the help of classical cryptography.

So that's basically like the three
main approaches, and each comes

with advantages and disadvantages.

Of course, there is no
silver bullet, unfortunately,

Dan: as is always the case yeah I
love the fact that you're describing.

I assume it's like changing the basis
state pretty much or the basis management

measurement for every single qubit,
which is sent across the network.

Now, can that also be done with sending?

Yeah.

Is there any way to obfuscate in
that way when you're sending the

instructions in classical information?

I guess there probably isn't, right?

No, because you...

Harold: I understand,

Dan: I think the way I understood it
was you were describing sending the

qubits across the network to then be.

To be received and then put onto the
computing processor and then perform the

processing and then return the results and
then you de encode based on your basis

state measurements on a per qubit basis.

Yeah.

But when it comes to executing quantum,
I mean, there's no way of doing that

currently with current quantum computers,
remote quantum computing because we're

just not there with with networking yet.

So.

Uh, Currently, anything that you execute
remotely has to be sent classically.

Is there a similar approach to
obfuscating what you want to execute

if you're sending it classically?

I would assume not, because
you have to send the...

Yeah.

Sorry.

Go ahead,

Harold: yeah.

Okay.

So let's go back to your first
assumption that says, okay.

Yeah.

In, in, in quantum networking,
we are not there yet.

Yes, that's true.

But at the same time we
are, we're working on it.

And um, so there, there is a, a very
large uh, European funded project

that is really tacking this question.

The idea is, is constructing like a
long range backbone across Europe.

And metro metropolitan networks connected
through a hub and within this this smaller

networks we are, I mean, very intensely
discussing with experimentalists to

try to co design a protocol that would
be suitable for very early networks.

Okay, so like it's more
like proof of concept.

But the idea is to have all the
ingredients of a full fledged protocol

working and scalable protocol to
be working on experimental nodes.

Okay so it's in the works.

Okay, it's not going to be like millions
of qubits sent around, but uh, but

it's really like putting everything in
place within the protocol so that if

you give me just more qubit and longer
coherence time, then I can perform just

the same protocol for a longer time,
which means like a bigger computation.

That's that's where we are right
now in building the networks.

Now coming back to really like,
can you do something classically?

Uh, well, Yes and no.

The very good advantage
in terms of overhead.

Of the first solution, which is like
sending qubits back and forth is

like, yeah, of course you need this
ability to send qubit back and forth.

But at the same time on the server
there is basically no space overhead.

Okay, maybe you are going to repeat
your computation a few times or

a few times or several times.

To get the result that you could
view it as a sort of, necessary

error mitigation in this.

Okay.

You can do this.

But security will not require you
more qubits on the server side.

That's one good point.

Because if you have like limited
machines, you don't want to actually

dedicate most of your computing
power to take care of security.

Because if you do this most of the
time, then what's, what you are

left with is a computation that...

You can do on a piece of
paper just by yourself.

Okay.

So if you have three qubits, they
are very well protected, but you

can't do anything with three qubits.

So that's one good advantage for the
solution of sending qubits back and forth.

If you have more powerful machines,
then what you can do is actually

have classical crypto come into help.

And basically what you are going
to do is having an encryption

of your preparation of states.

Being sent classically to the quantum
computer and because the quantum computer

cannot break these things encryption,
what you are really performing in

terms of actually quantum operations
will still remain understandable

from the point of view of the server.

Okay.

But of course, when I'm saying
this, it's like you still see

what the server is doing, right?

I mean, You can check and track back
all the instructions that you sent.

They are encrypted.

You don't know what it is for.

But you can still track them, which
means that if this qubit is going to

be secure, okay, if the aim of that
thought is to produce the same qubit

that you were sending across the network
it needs to be operation enough and

complex enough uh, quantum operation so
that you cannot simulate it classically.

So now comes into play the okay.

Where is the quantum
advantage in the story?

Uh, In the sense, like, when
do you estimate that you have?

A quantum operation that is complex
enough so that the classical computer

cannot break it or cannot simulate it.

And usually what you end up with
is an overhead in terms of space

in this time that say at the very
minimum will be like 100 times.

Okay, so for having one, one
protected qubit, you need to

have like 100 physical qubits or.

Yeah, 100 cubits unprotected.

But yeah, these needs to be qubits
that are very good quality, okay,

because if you can, if you start
adding noise to this well, then

it's going to be a bit complicated.

So it's actually 100 fault
tolerantly protected cubits.

give you one fault
tolerant and secure qubit.

Okay.

So it's quite an overhead.

And so right now, I think we need to
push in both directions and try to

simplify the protocols as much as we can.

Maybe instead of having 100 qubits uh,
overhead, we can have like, yeah, sure.

100 qubits.

Per qubit produced, but if you
produce them serially, maybe you

can recycle this overhead and, yeah.

So these are the kind of questions
that are very important to address

to try to simplify these protocols
and bring them to a regime where

they can actually be implemented.

Right now, these classical crypto
protocols are, most of them are.

Completely out of of reach
but yeah, doesn't mean that we

shouldn't do some effort to try to
simplify that as much as we can.

Dan: It's fascinating talking about
the need for additional qubits in

that scenario to secure the qubit.

But then you also need additional qubits
to create a fault tolerant physical qubit.

So a logical qubit, sorry.

Yeah, I mean, it just grows, doesn't it?

It grows and grows, and it's going to make
it uh, you know, as long as we're in the

NISQ era, it's going to be very difficult
to perform this type of operation.

Unless you're doing it on the fly, sending
the state of the Qubit across the network.

So, Yeah, that leads us on to networks.

What um, we could go in lots
of different areas here.

What research are you doing when
it comes to networks, or which

organizations are you working with?

What's your main focus when it comes to...

To networking.

You mentioned the European I don't
know if that was the Euro QCI that

you mentioned or something different.

Harold: Actually, I mentioned most, I
mean, the European project that we are

part of is the Quantum Internet Alliance.

And it's...

It's, yeah I think most of the
people doing uh, quantum networks

are part of this of this project.

It's really huge.

I think there are more than 40
partners companies and and academics.

And the idea there, I mean, what
we try to do is is represent,

let's say, the end user.

Okay, so we are the computer
scientists the theorists and we

try to say okay Here is a protocol.

What can you implement from this?

What kind of physical parameters can
you give us then we'll try to go back

and simulate that understand what will
be the performance of the protocol?

Basically, it's these kind of protocols
are I mean, it's it's very interesting

because when you are looking at
verification of quantum computation, the

model is the following you have a client.

You have a server.

The client is honest.

So he's doing everything like
honestly, perfectly following the

protocol and things like this.

And then you have the server and the
server can be arbitrarily malicious.

Okay.

And basically the server starts
or the malicious party starts.

Okay.

Really at the door of
the lab of the client.

Okay.

So as soon as something
leaves The lab of the client.

We are thinking that it's maliciously
controlled Okay, so even if you have

some noise the genuine noise that
happens we think okay, that might be

malicious Okay so it what happens is
that if you manage to certify in this

condition that the Computation has been
performed correctly In fact, you have

been also saying the noise, imagine
now that, that, okay, the malicious

party decided not to be malicious.

Okay.

But it's still noisy.

Okay.

But from my verification
perspective, I don't know this.

Okay.

I don't know that he's being
honest, but so I'm not changing

anything in my protocol.

But if I still verify that, that means
that the noise that has been happening

outside of the lab of the client, okay.

was not very strong because I was still
able to guarantee that the computation

was performed correctly, okay?

In spite of the noise,
I still have my result.

So in fact, these protocols
are surprisingly interesting

for, as a benchmark, okay?

Because in the end, if there is
no malicious party, what we do is

that we are benchmarking the noise.

So people doing experiments are very
happy with this kind of protocols

because it's like the ultimate benchmark.

Okay, because if I pass the
test, I can guarantee that their

experiment is running well.

And it's very rare that you have this.

Okay, because when you do an
experiment in the lab, you always

have assumptions and things like this.

Okay, and we are just providing a
methodology saying okay, I don't

want to know about your assumptions.

I don't care.

If you pass this test, Then I can
give you a cryptographic guarantee

that you've done the job properly.

So that's very nice.

And basically, we are working
with them to make sure that they

don't break the protocol as well.

Okay, because still the protocol
needs to be done in a certain way.

Okay, most of the time when you are doing
or running an experiment in the lab.

The client is here, maybe two meters away,
there is the server and maybe there is

like 50 kilometers of fiber in between
them, but they are still in the same lab.

So you don't want to make, you want to
make sure that there is no communication

that that happens between the client
and the server there in this situation.

So they need to close this,
this kind of loopholes.

And we need to make sure that by
processing the data and things like

this, they don't introduce problems
in the, in the protocol but, but,

but if you do things properly,
then it's a super nice benchmark.

And then from there, what we do in
fact is try to help the people from

the network assess the performance
with respect to this benchmark.

And we try to work.

In tandem with the various
hardware labs involved.

Okay, what can you do?

What can you improve?

Maybe they have devices that we didn't
think about that can be incorporated

in the protocol and just improve the
robustness of the protocol by itself.

Okay.

But then we need to change the proof uh,
to make sure that everything works again.

But we like to do this.

So it's okay.

Dan: Sounds fascinating.

What could you explain the
verification methodology?

For me, in terms of which level
it operates, is it a series of

steps which need to be followed?

Is it a series of very specific
tests and variables that need

to be captured and measured?

Harold: Yeah, it's very simple.

It's very simple.

Just before in the discussion, I convinced
you that it was not very complicated to

hide the computation and the data, right?

So by just changing, having this,
this relative reference frame that

that I'm using to communicate with
the server, but keeping my own private

reference frame secret and varying
this reference frame qubit by qubit.

Okay.

So let's build on top of this.

Okay.

So now I'm starting in a situation
where I'm able to delegate

my computation to the server.

And he doesn't know what I'm doing.

And now I want to verify
that he's not doing something

malicious to my computation.

Which means I want to have a
guarantee that he's following the

instructions that I'm sending to him.

The problem is that let's say I'm doing
machine learning I'm sending these

instructions to the server, and maybe
he's doing something else, and still,

the data are going to look okay to me.

Or they are maybe going to
solve most of my problem.

But the but the server has nonetheless
tampered with the actual real solution.

And what I'm ending up with is maybe
something that's not bad, but it's

not the optimal that I paid for.

Okay?

And I don't want to be in this situation.

The problem is like, if I'm actually
doing this computation on a quantum

computer, that's precisely because I
cannot do this computation on my own.

So I cannot check.

Okay.

But then if you are able to perform
computation in a blind way so that

the server doesn't know what you are
doing, maybe from times to times,

just ask him to do a computation
for which you know the result.

See because you delegate something
that is blind, it doesn't know.

Okay.

But it's a test.

And if he fails the test, you know
that he's doing something nasty with

your computation, and that's the idea.

It's very simple.

You just and then you just need to work
out the math and it's working nicely.

So basically, you do that several times.

And you do this.

And so what in the end, what we
do, we want to be a little bit more

technical is basically in these
blind computations, you can delegate

blindly full classes of computation.

So there you have a class of computation
that can be delegated blindly.

So every computation within this class
will look the same to the server.

And within this class, we know that there
is always a large portion of this of

these computations or big enough portion
of these computations that are Clifford

computations and Clifford computations
have the nice property of being

Super easily classically simulatable.

Okay, so you can simulate
that on a classical computer.

It takes like milliseconds.

Okay Even though you are using like very
large computations Okay, so you can do

this classical computation on your side
as a client you blind it You send it to

the server and you ask for the result
and then you compare what you got and

yeah, so So that's the way it works.

What really is important is...

is to actually prove that this
class of Clifford computation is

large enough to catch every possible
deviant behavior from the server.

Okay, but if you prove this, then
basically other and then you can

devise, like Which kind of pre for
the operation you want to take?

What is the best one?

What are the ones that you can
actually easily implement because

you have restrictions on the
devices and things like this.

And then you have like plenty of
questions that come after this.

But the principle is very simple.

Dan: I tell you, it's a fascinating
field that you're working in, right?

Because the networking at the moment
is still at the experimental stage, of

course, and we're already thinking about
ways to make these executions of quantum

instructions over the network private,
which I think is, it's a fascinating

field, and I'm sure what you're doing
will become very central to the way

quantum computers are used in the future.

I think I'm going to, I'm going to
wrap it there because it's been such

an amazing talk and I think we should
probably reconnect in a year or so

to get an update for how, how some of
these projects have been going and maybe

how you're thinking is is evolving,
but that was absolutely fantastic.

Thank you very much for
the discussion, Harold.

It was

Harold: a pleasure on my side.

Dan: Good.

Yes.

Same for me.

Okay.

Bye for now.

I'd like to take this moment to thank
you for listening to the podcast.

Quantum networking is such a broad domain
especially considering the breadth of

quantum physics and quantum computing all
as an undercurrent easily to get sucked

into So much is still in the research
realm which can make it really tough for

a curious IT guy to know where to start.

So hit subscribe or follow me on your
podcast platform and I'll do my best

to bring you more prevalent topics
in the world of quantum networking.

Spread the word.

It would really help us out.