851: Quantum ML: Real-World Applications Today, with Dr. Florian Neukart
01:11:51 Share

This is episode number 851 with Dr. Florian Neukart, Chief Product Officer at TerraQuantum.

Welcome to the Super Data Science Podcast, the most listened to podcast in the data science industry. Each week, we bring you fun and inspiring people and ideas exploring the cutting edge of machine learning, AI, and related technologies that are transforming our world for the better. I'm your host, John Krohn. Thanks for joining me today. And now, let's make the complex simple.

Welcome back to the Super Data Science Podcast to kick off our first guest episode of the year. I'm delighted to have today's exceptional episode on quantum machine learning for you with one of the best people on the planet to fill us in on quantum ML. That's Dr. Florian Neukott.

Florian is Chief Product Officer and member of the board at TerraQuantum, a leading quantum computing startup headquartered in Switzerland and Germany. He's also Assistant Professor of Quantum Computing at Leiden University in the Netherlands. He holds a PhD in quantum computing and machine learning. Today's episode gets a bit technical at some parts, particularly near the beginning, with respect to the mechanics of quantum computing and quantum ML, but by and large, the episode should be fascinating to any interested listener.

In today's episode, Florian details how a new generation of hybrid quantum classical systems has made quantum computing practical for real-world applications. He provides an overview of the available quantum computing chips, including Willow, Google's new quantum chip, which has been making a big splash. He talks about the race to develop quantum

quantum proof encryption, the breadth of real world problems that can be tackled with quantum machine learning today, how quantum ML could unlock personalized medicine, nuclear fusion energy and revolutionary space technologies, and how you can get started with quantum ML yourself today. All right, you ready for this exceptional episode? Let's go. Florian, welcome to the super data science podcast. I'm delighted to have you here. Where are you calling in from today?

Thanks for having me. I'm calling in from San Francisco. Nice. And we met in person in Lisbon, Portugal at a web summit. We met backstage. You gave a great presentation on quantum computing, quantum applications, how this space is moving. And I knew instantly that I wanted to do an episode about this. We had an episode a couple of years ago

on quantum machine learning. So if people are interested in that, they can go back to episode number 721. It's with someone named Amira Abbas. Have you come across that person, Florian? Yes, of course. I know the community is still fairly small, so the quantum computing folks tend to know each other at least by name. I know her, yes? By name. It was a great interview, and it was interesting because...

She poured cold water on the idea of practical quantum machine learning applications being useful in the near term. Is that something that you'd say, you know, that's a safe statement to make? No, I would disagree here. But it depends on how you look at it. So when you think about

Basically how everyone started doing quantum computing was everyone thought about how can I translate a real-world problem into something that can be processed on a quantum chip.

And then we tried to reduce complexity. We made the problem smaller. So in the end, it wasn't a real world problem. You would just solve a toy problem. But then people realized, well, we have all these other compute power out there. Hybrid is actually the way to go. So you take all the classical non-quantum high performance computers that you have and use them. They're still good at what they are used for.

And then you add quantum computing, but only when necessary. And that's what we do in quantum machine learning too. We don't take, for example, an entire neural network and try to express it as something that can be run on a quantum circuit, on a quantum chip. We take only parts of it. And if you do it that way, then quantum machine learning, hybridized way, can be useful today already. Nice. Well, that's exciting. So I guess that kind of hybrid approach is something that we are going to be talking about a lot today.

In fact, you guys at Terra specialize a lot in taking advantage of traditional computing and kind of allowing us to have simulations of quantum computing, that kind of thing. I guess maybe explain a bit about Terra Quantum and how it fits into the broader quantum picture. So Terra Quantum is a quantum technology company. We focus on all pillars of quantum technology in both hard and software development.

Now, it always depends on who you talk to when you ask about what are the pillars of quantum technology. For us, it's quantum computing, it's sensing, it's imaging, it is cryptography. When we talk about hybrid quantum computing, so the

idea here is really develop solutions develop libraries that efficiently combine all the classical high performance computing plus quantum computers now the terminology sometimes may be misleading you so everyone in the in the in the field currently uses hybrid and we all talk about integration but then sometimes when you look at it you find it's

It's just a cloud provider that offers all the classical HPC that we all know so well, and all of a sudden an additional component, quantum processors. But how you combine it, how you efficiently leverage this new power, that is then up to the user. So for us, it's different.

we think about integration and bringing quantum chips into our libraries. So think about maybe one part of a deep neural network, a fully connected layer, for example,

that you express as a quantum circuit. Then a quantum circuit, by the way, is a set of gates that you stack together. And that is then generally how you express anything that you run on a universal or gate model quantum computer. Hence the name gate model quantum computer.

Now, you express that neural network in that form or this part of the neural network, then you still have to bring in the device. So the quantum device may use other gates. So they may not be the same as we have in our software. So we have to do a mapping. There's integration into the software is a certain effort, but we do that. So it's not that we want the end user to worry about that. For the end user, in our view, the only thing you should worry about is what quantum chip I want to use.

No matter the technology, no matter the topology of the chip, they should just switch a parameter. I want to use quantum chip A. It's a different parameter than using that other quantum chip B. And you can compare the performance. Still, the integration effort is on us. You have to do it.

It's not something that can easily be done, but that's how we offer it and that's how we see hybrid quantum software. - I have a couple of questions from what you just said. One of them is, you said gate in there. So what does that mean, a gate in this context?

So when you think about all classical computing, so non-quantum computing, then the most fundamental level, we also use gates. We have the AND gate, the OR gate, XOR, NOT, and in the end we want to have a set of universal gates that can do any operation that we need. Now, the same thing we do in quantum computing. The difference is that because of how these gates are designed, there is in theory an infinite number of gates.

We of course don't want to have an infinite number of gates, so we still limit it to certain gates that we use very often and efficiently to change the states of a quantum bit. So we do the same thing as we do in classical computing, but it's different because we're using quantum effects. So we have all this richness of quantum information where one bit can not only be in one state at a time, it can assume a superposition of states

which gives you much richer compute capabilities. So in that case, we apply still these gates on our quantum bits.

and apply a sequence of it, then you call that quantum circuit. And that's the algorithm. So the whole artistry in quantum computing is very often to find the right set of gates, combine them efficiently, then also think about the properties of the device. Because these systems are very delicate. I cannot have an algorithm run for minutes. You only have nanoseconds, maybe microseconds on the chip.

So it's a very small amount of time that I have to execute that circuit on the chip. So I must be very mindful how deep the circuit is. The more I interact with it, the more likely I mess the system up. So that's what we prepare in our algorithms.

predefined sets of gates mapped to specific devices so an end user can run a hybrid quantum neural network and manipulate the layer still, manipulate the architecture, but doesn't have to worry about how to really access the device.

Very interesting. And so the different kinds of devices, you talked about how your Terra software allows you to have the particular gates that you'd like on a given kind of chip. An interesting tidbit that you threw in there is that you only have very small amounts of time to run any kind of compute on the chip, which is something quite different from classical computing. You can leave your classical computer running for months, sometimes without any issues, and

Whereas you said they're kind of nanoseconds or microseconds. So that's how quantum computing works all the time these days? Yes, that's the challenge. So you really want to be fast in the execution of your circuit. There are different things to consider too. So if you have, you talk about coherence time. So this is,

the time that the quantum computer can hold this very delicate superposition. So all possible configurations of bits at once. So if you formulate the algorithm, then that would mean all possible solutions to a problem at one time. And when you make the system collapse, when you read out, so to say, then the whole artistry is getting the right solution out of all the many billions or sometimes trillions of solutions that you have available at a given time.

So that's what we have to worry about. And then thinking about the depth of a circuit. So the deeper a circuit is,

the more likely it is that the quantum computer, the system, collapses before the algorithm is completed. So ideally, I can limit my operations, don't have two deep circuits. I can limit them in terms of the number of qubits that I access at once. So for example,

have a single qubit gate operate on the chip that causes less harm, so to say, than having an operation that goes over many qubits. It's harm in the sense of risking collapse of that system. Gotcha, gotcha, gotcha. And something that I'm familiar with from episodes like 721 with Amir Abbas is that a qubit is the quantum computing equivalent of a bit, but it has different properties than our classical bits, right?

Exactly, yes. So still, when you think about the classical bits, and that's also to what you said before with a classical computer, I can run it for a month. I can even stop computation, read out an intermediate result and then continue. This is all challenging with quantum computers and quantum bits.

So still, as you said, we have the smallest unit of information, the quantum bit in a quantum computer, but it's different. So think about maybe an electron. Let's take the simplest atom that there is, a hydrogen atom. We have one electron in the orbit and then we look at the electron only that has one quantum property that is called spin and the spin can be up and down.

And interestingly, as long as I don't look, and in that case it means interacting with polarized laser light, as long as I don't look, the

states coexist. So it's up and down at once. And that's where the remarkable power or one of the reasons why quantum computers are so powerful. So imagine you have two bits. So each of these two bits in the end will give you a zero or one. But as long as I don't look, it's zero and one at once. It means two to the power of two equals four possible configuration that that system can assume at once.

If I have 3 bits now, I have 2^3 = 8. So this is very remarkable. It says that in a perfect quantum computer, anytime I add one quantum bit, I double its computational power. If I have 1000 or 1001, there's a significant difference.

Now, in more practical terms, that means if I have expressed my solution in a way that I can embed it on a quantum chip, then all possible solutions to that problem coexist. And now what a quantum algorithm does usually is it makes all the solutions that you don't want unlikely to appear when you look at the system, when you measure it.

They're never gone. So quantum computers are probabilistic systems. If you do everything correct in your algorithm, you can still get a nonsense solution. So that's why you don't only measure the system once, you do it a thousand times. You're still very fast, still can be done in microseconds.

And then you get 800 times agreeing solutions, 200 times some random solutions. So then you would do a majority vote and say that's the correct solution to the problem. And there is entanglement too. So I don't want to make this a physics lecture, but there are more quantum effects at play there.

As a super data science listener, you're probably interested not only in data-powered capabilities like ML and AI models, but also interested in the underlying data themselves. If so, check out Data Citizens Dialogues, a forward-thinking podcast brought to you by the folks over at Colibra, a leading data intelligence platform.

On that show, you'll hear firsthand from industry titans, innovators, and executives from some of the world's largest companies such as Databricks, Adobe, and Deloitte as they dive into the hottest topics in data. You'll get insight into broad topics like data governance and data sharing, as well as answers to specific nuanced questions like how do we ensure data readability at a global scale?

For folks interested in data quality, data governance, and data intelligence, I found Data Citizens Dialogues to be a solid complement to this podcast because those aren't topics I tend to dig into on this show. So while data may be shaping our world, Data Citizens Dialogues is shaping the conversation. Follow Data Citizens Dialogues on Apple, Spotify, YouTube, or wherever you get your podcasts.

Yeah, and quantum entanglement ends up being the thing that in yoga studios and stuff, then that it ends up being explained for how minds talk to each other, which is like such an interesting social phenomenon that's happened. And I bridge both worlds, like this kind of science technical world. And I also, you know, I like yoga and, you know, I like, you know, meeting, you know, these interesting kinds of people who in a lot of ways are living, you know, a great life. But it's a funny thing to see how

You see big authors like Deepak Chopra take things like quantum entanglement and use them as proof, as an explanation that minds are talking to each other and we can, I don't know, be accessing past lives or other times or I don't know. It's interesting, but it's not as exciting as all that. Although it is, I mean, it's a really interesting phenomenon.

but it is yeah yeah it is so i mean in the end i i really encourage this uh discussion over different branches of science and philosophy because in the end what physics very often has become is

You make an experiment, you describe the experiment with an equation, but sometimes physicists forget to ask what does it mean? So for example, if you think about superposition, this coexistence of all states at once,

depending on what physical world you have, but it could mean the world is probabilistic and reality doesn't exist unless I look, unless I measure. So what does that mean? That's a question that physics, I think, is not very good at answering. So that's why I think this discussion over, or

or with other branches. It's a very, very important one. Yeah, it is mind-bending. And I think even Einstein has famous quotes about how quantum theory, he was like, this can't be right. It doesn't make any sense. Yeah, made him a little uneasy, yes.

And especially entanglement, he didn't like too much. So I'm sure everyone in quantum physics knows it, but he called it the spooky action at a distance and he was not convinced. Right. Well, I mean, now we have real quantum applications happening. So yeah, tell us a bit, you know, now we have a bit of a sense of the theory and the special things that you can do with quantum computing and

So can you provide an example of a practical, maybe optimization problem? That seems like the kind of thing that you guys do at Guterra Quantum a bit. So like some kind of practical problem that is intractable for classical computers, but with some quantum computing as well. It sounds like typically a hybrid system.

How we can have a real-world application that provides some value. Yes. So there are so many. So these three branches that we look into, everyone who does quantum computing does, is machine learning, and as you said, optimization, and then simulation. One problem in optimization that sounds boring at first is scheduling.

But that is impossible to tackle with no matter how powerful a classical computer you have. So the challenge is manifold. Scheduling appears in production, scheduling appears in hospitals when you have to do plans for the nurses and the doctors. Scheduling appears in computers in electric vehicles when you want to optimize the subroutines for power consumption.

One of the things that we did with an automotive company, with Volkswagen in that case, was a scheduling problem for production. So imagine you have vehicles coming out of the production line, then all of these vehicles must undergo a couple of tests.

Ideally, I can test every vehicle for everything. But the reality is you don't have enough time, you don't have enough people, and not all of these people doing vehicle tests have the same skills. Especially if it's emissions testing. I mean, then you've really got to skip a few cars.

Yeah, that one. So some of these tests, of course, you can plan because you get reports, field errors. The workshops will report, well, I have these couple customers complaining about water damage. So anytime it rains, it gets wet inside the vehicle. So then you do water tests. But then there are 250 something test classes and each of these test classes has subtests. So

So the question now is, given the staff, the personnel in production with the skills available today, how can I maximize the number of tests for all of these vehicles that come out? And that is a very complex scheduling problem. But the same algorithm can be applied, as I said before, for scheduling subroutines in vehicles.

in electric vehicles, so you want to minimize power consumption, so then maybe you have two subroutines that use the same data. So instead of loading into memory, deleting it and loading it again, maybe I can execute these subroutines in sequence and access the data in sequence before I delete it. So these are things where this can be applied.

which don't sound very exciting at the beginning and you would wonder if there's really something where I would need quantum computing, but you do because in the end with classical non-quantum algorithms, the only thing you can do is heuristics and make approximations. So you can never be sure is this really the best solution I can find.

I must admit, also with a quantum computer, you cannot be sure. But what you can do then is you just compare the classical and the quantum algorithm. And if the quantum algorithm gives me a better solution, then that's the one.

that I take. Other problems are in logistics. We did many logistics optimization problems. So imagine you have a fleet of vehicles that have to transport goods through a network of hubs. So for example, food which can decay, you have to have that vehicle number one at a certain hub between 1:00 and 3:00 PM. Otherwise, there is a problem with the food, for example.

So how do you optimize the number of vehicles that I have in my transportation fleet, minimize the number of vehicles that I need to transport all the goods efficiently through the network? Or in other ways, how do I reduce the empty miles? The empty miles meaning I have trucks that just go from A to B but don't have any load. So how do I avoid that? So this is also one of the things, one of the problems that we have solved with a customer.

And then it ranges from optimization of satellite constellations, which we did, financial optimization. So you want to predict market behavior, you want to do collateral optimization, you want to do exotic options pricing.

You want to do machine learning. You want to learn better, do better image classification. So all of these things benefit from hybrid quantum computing. Interesting. And so let's dig a bit more into that machine learning one that you mentioned there. And you said there were three types, three areas of quantum application. You said machine learning, optimization, and I didn't quite get the third one down. That was simulation. Simulation.

Yes. Usually when people in quantum computing talk about simulation, they mean the simulation of physics and chemistry. So imagine on a most fundamental level, you have one quantum system, the quantum computer that you use to simulate another quantum system, the molecule.

And what you want to do is, for example, invent better battery chemistry. You want to design better cathodes or anodes for batteries. You want to make batteries smaller, be able to charge them faster. You want to be able to charge them more often before they degrade. These are things that require quantum simulation. You have to be able to understand the molecules involved in all the electrochemistry.

And the same is true for finding drugs. If you want to understand how a drug affects a protein in your body, I have to do it quantum mechanically. Right now, the drug design process is different. It's experimental. You design and conduct an experiment. You do it with lab animals. But if I now were able to do it with a quantum computer, so really effectively basically simulate through all the possible atomic configurations, all the molecules, I could do

do that with a quantum computer and then only make one experiment because I would get the right drug out right away. So that being said, simulation is the most complex problem for quantum computers. And while there has been tremendous progress also in TerraQuantum, there is much more that's going to happen with more powerful systems. Very cool. I love it. A while ago, when I asked you about gates, I got thinking then about hardware. You were mentioning

different kinds of chips, you'd say, you know, TerraQuantum solution allows you to have whatever gates you need on chip A, chip B. What are the kinds of key chips that people use today for quantum computing? And if

This is, we've also, we've got to talk about this Willow chip from Google that made a huge splash recently at the time of recording. So yeah, so fill us in kind of on the quantum hardware scene and how you might, you know, a client comes to you and says, hey, you know, I want to do this computer vision problem or I want to do this logistics problem. You know, how do you say, you know, let's try using this chip first and here's why.

Yeah. So in the end, I think there is not one answer to that question. It depends on the customer. It depends on the existing contracts that they have. So, for example, IBM is very advanced with their quantum chips. There's Rigetti. There's D-Wave. Different paradigm, though. There is, of course, Google.

But then sometimes, you know, companies have contracts already, cloud contracts. So for example, have a cloud contract with AWS, then whatever AWS has to offer in terms of quantum chips, that will be easily available.

People have contracts with IBM in whatever cloud systems, high performance computing. So then of course, IBM may be the preferred vendor. But it really depends. I think there is not one answer to that. It depends on the preference. So that's why we develop software that doesn't care about the quantum chip. For us, really what a customer wants to use, we make available. And if we want to give them the ability to compare to other systems,

when easily done. So we can do that. It's interesting. IBM is this quantum computing seems to be one of the few places that IBM is still near the cutting edge. There's so many other things that they're kind of a byword for a past age. But with quantum, they are, you know, they're playing right at the forefront. Yeah, I can't speak much about the other areas, but they are fantastic research going on.

And they do a lot of work in error correction too. So that was very important. Recently released a couple of good papers. So it's really, I would say all of these institutions working on these systems contribute to progressing the field so much. It's of course strongly dependent on the software, but then without quantum systems,

hybrid quantum software wouldn't be able to grow either. So therefore, every contribution that any of these institutions make, any new system that we have, will make the software more powerful. So for us, imagine you would use our software stack and use an algorithm, say a hybrid neural network to solve a problem. Then once the quantum chips become more powerful, our software is designed such that

that you just underneath plug in the new system, make it available through API or whatever, and the software becomes more powerful too. There's no need to redevelop the algorithm. There's no need to worry for a customer about this new system. We worry about that. But in the end, more powerful chips mean the algorithms become more powerful too. So it's a good thing that many institutions work on it. And so more powerful usually means more qubits? Yes.

More qubits, better quality qubits, better coherence time. So the coherence time is one of the things. So you want to be able to run more complex algorithms, which means I have to have the system be able to maintain their quantum states for a longer time. So that is a question too. That's also, by the way, one of the research areas that we conduct. So we develop qubits. We have a new design for qubits.

And the goal is to increase the temperature. So at some point reach room temperature at superconducting qubits two and then still have better aero behavior, better isolation towards their own irrelevant physical properties and the environment.

But that's what everyone is looking at and everyone is working on, no matter what technology it is. So, yes, it means more qubits, better connectivity too. So if you think about these quantum chips, when you look at them at a 2D plane, you have your qubits and then you have physical connections between these qubits. In an ideal world, I would have

all to all connections, but you cannot build that. There's too much interference, there's too much complexity. So you have to think about what is the right amount of qubits that I have to connect to each of the neighboring qubits? How is the optimal topology? So that alone is a very complex problem.

And if you think about it, the more or the higher the connectivity, the more complex problems I can embed on a chip with less qubits, fewer qubits. So the worse the connectivity is, the more qubits I need. So all of these things play together. Nice. Very interesting. Going back a tiny little bit,

you were talking about how IBM has done a lot of work on reducing errors and that going back a little bit more to my question about Willow, this chip from Google, is that that seemed to be kind of the main innovation is that Willow, so I can't go much beyond the blog post level on quantum computing, but so this individual named Hamot Nevin,

who's founder and lead of Google Quantum AI. And he says, I'm delighted to announce Willow, our latest quantum chip. And one of the big achievements here is that Willow can reduce errors exponentially as we scale up using more qubits. And so the implication there to me is that

Historically, having more qubits would mean more errors. And somehow they've figured things out here so that your errors actually reduce and they reduce exponentially apparently as the number of qubits increases.

Yes, that is true. So historically, it's just as you said, imagine you have, so how errors are measured, there are different ways to do it. But one way to do it is fidelity, two qubit fidelity. That means you look at a two qubit operation and apply that to two qubits. How often do you get the expected results? Then you get some value, maybe 99.7% or whatever, very high.

So it sounds like it's already near perfect. We should be fine. But then we aren't. Because if we scale that system up to millions of qubits, then the error would multiply, so to say. So you would, with millions of qubits, even with that high two-qubit fidelity, still get random results. So this is no good. What Google now showed was that they can't correct for these errors and they can't scale it up. So that is fantastic news.

for the whole, the entire field. So that means the barrier is not in the air anymore. There are some other challenges. Of course, you have to worry about engineering challenges. The more qubits you have, say millions, the warmer the system becomes. So you have more intense cooling requirements if it's a superconducting system.

and some other things. But what Google showed, and this is the remarkable result, is exactly as you said, they can scale the system up, grow the system, keep the error down. That is one of the things that's needed for having bigger, better chips. So very remarkable. Ready to take your knowledge in machine learning and AI to the next level? Join Super Data Science and access an ever-growing library of over 40 courses and 200 hours of content.

From beginners to advanced professionals, Super Data Science has tailored programs just for you, including content on large language models, gradient boosting, and AI. With 17 unique career paths to help you navigate the courses, you will stay focused on your goal. Whether you aim to become a machine learning engineer, a generative AI expert, or simply add data skills to your career, Super Data Science has you covered. Start your 14-day free trial today at superdatascience.com.

That's fascinating. And it provides us some more context on the kinds of the nitty gritty, the detail on how these things work. Let's move back to applications. We kind of got going there on applications a while ago. You talked about in machine learning, optimization and simulation as being these areas of quantum application. What are the key challenges you face scaling applications?

quantum approaches to meet what must be the increasing demands of enterprises. So as one financial firm figures out that they can be optimizing better with a quantum chip, then every other financial firm is going to want to jump in and be using quantum chips as well. So yeah, what are the key challenges and how does TerraQuantum address these challenges through cloud-based and hybrid computing approaches? So

So that's a very good point. So with these use cases that we publish, so usually what we do is

Two things. So we not only develop a proof of concept, we go in with the intention to use whatever we develop productively. And then we compare to best in class or best in business algorithms. So it's not that we just want to sell the quantum magic, so to say. For us, it's really important that it matters practically. If you have an optimization problem that you currently solve with Cplex or Gurobi, that's what we want to beat.

And that's our goal and promise when we go into a use case. I say, well, whatever you're using, we will be better. And if it's not better, then from a customer perspective, it doesn't make sense to use it, no matter if Quantum is in it or not.

So and that because of these publications, because of the studies that we release with customers, the use cases that we do, of course, more and more customers get excited or potential customers get excited about quantum computing. And that's good for the business. Good for us. One of the challenges that we still see, though, is that there is not the one solution that fits all.

So what we try to do with our software development is really develop generalized or generalized stack that you can use to solve arbitrary machine learning problems, arbitrary optimization problems.

It's still growing. So sometimes we go in with a customer and then they have a problem that we cannot solve with our current stack of algorithms. So we have to research one and develop one, invent it basically. And that's, I mean, it's a lot of fun. But if you,

focus on only that, you can only do so much as a company. So therefore, it's always this balance between finding the right existing solutions that we have and applying them versus doing research and inventing something completely new for a new customer. And I think we're getting better at that. But still, there's a lot of work to do. And when you think about the classical optimization fields or the non-quantum optimization field,

I don't know how many hundreds of thousands of contributors over the decades there were to that field. And this is what's happening now in quantum computing too, in all the areas that everyone is looking into. However, the quantum community is much smaller and much younger. So there's still a lot to do. And then, of course, the next question is,

How do you run it? So still many companies, it's not only old economy companies, many companies, they say, I can only run this locally. I don't want to have anything in the cloud. Now that's a challenge because the quantum chip is in the cloud. So

So usually it doesn't happen that someone has a quantum chip in their local data center. Systems like superconducting systems, because of the fridge, because of the cooling requirement, they cost you 15, 20 million. And next year they're old because it's not something that you would locally get. So then you have to find ways around that. So how do you, for example, prepare the data? So that's also something that's very special to quantum computing. You embed data.

data on the chip. So the chip is memory plus processing unit. And embedding means I have to translate my problem into zeros and ones. Now, if I don't have the algorithm that does that translation, then I have no clue, no way, even if I'm an attacker, what that problem is that I'm submitting here. So that poses sometimes less risk. So even if you were to submit it unencrypted to the cloud, which no one does anyways, you

it helps us to get the worries or at least reduce the worries on customer side. So if you transmit a list of zeros and ones, can be the most complex optimization problem or a recipe for spaghetti, so to say, no one knows, except they know how you prepare your data.

But that's something you have to explain. It's not that people know about all that stuff. So it's bringing in new technology. Now it's getting enterprise ready. It is enterprise ready. People are learning about it. And that's a journey that we do together with our customers.

Very cool. Another journey that you've been doing together with somebody is I've been reading about collaborations with NVIDIA, which is obviously a topical company to our listeners, one of the most valuable companies by market capitalization in the world, developing chips that have ended up being super useful for AI applications. And so tell us a bit about TerraQuantum's relationship with NVIDIA and how this is facilitating quantum AI

AI innovations? So the story here is the same as with many other backends. We make NVIDIA's backend, the hybrid quantum cloud, available through our own platform. So we try to be as agnostic as we can be. Whatever provider a customer wants to use, whatever cloud system they want to use, we make available. That's many reasons because of existing contracts. So they say they have a GCP contract. I only want to use GCP.

Alright, that's where our algorithms run. You can benefit from our existing contract. And that's the same reason for why we do this partnership and integration with NVIDIA. Plus, NVIDIA has fantastic hardware. So, and fantastic research as well in quantum computing, in simulating quantum systems.

And if you have that available, it can only be a benefit. And then there is generally in quantum computing this other approach that we, for example, strongly rely on, which is called tensor networks. And tensor networks is a purely classical algorithm at this point, but strongly relies on GPUs. Now, we already beat with classical tensor networks many state-of-the-art algorithms.

For example, recently, one of our packages that we released is TQChem. We focus on chemistry where we have the conformer search. So you want to find the optimal spatial orientation of molecules and how they fit together to do something, to treat, for example, a disease in the body. Now, the tensor networks that we use for that rely on GPUs, but

tensor networks can easily be translated into quantum circuits. So now if we have more powerful quantum systems, at some point we take the very same algorithm, just translate it into a quantum circuit and squeeze out even more performance. So that's the reason behind all that too.

Nice, nice. That's interesting. Yeah, and it does sound like you guys are providing a great platform for working with any of these kinds of backend options. NVIDIA is just one on the list. I wasn't aware until now of these kind of NVIDIA quantum efforts, so that was the new thing. Yeah, they're spending a lot of effort on it, yeah. So something that comes up a lot with quantum computing is

is that people are concerned about encryption being broken. And so, for example, Satya Nadella, the CEO of Microsoft, wrote in a book that it would take a classical computer a billion years to crack a widely used encryption algorithm, RSA 2048, whereas it would take a quantum computer less than two minutes to do the same. So a billion years versus two minutes. I also find it funny. I mean, this is the kind of thing that

With Willow, they're like, oh, we did this thing in five minutes that would take the world's fastest supercomputer 10 septillion years, a number that vastly exceeds the age of the universe. And anytime I read something like that, I feel like it's been, you know, it's like an unfair comparison because it's like you're...

You know, you're using a problem that is ideally suited to quantum computing that is unsuitable to classical computing. And so, you know, you get these big numbers and I guess it is kind of impressive. But also to me, every time I read those, it isn't that surprising or shocking because I already know that quantum computing can solve problems efficiently that classical computers can't. Anyway, encryption is one of those things.

And so in a world where encryption today on the internet and so many aspects of our lives is required for trust between strangers and ensures privacy,

you must have spent some time thinking about this. I don't know any of the answers, but people ask me. They say, what about quantum and how it's going to break all of our encryption records? And I give this kind of vague answer to people. I'm like, yeah, but it requires this one-upmanship. It's hackers versus...

people who are trying to come up with security solutions. And I don't think there's going to be some encryption apocalypse where all of a sudden everyone can have access to everyone's account. It just seems to me like there are going to be solutions, but I have no idea how those would work. So maybe you have some insight into how these future quantum proof encryption algorithms could work.

Yes. So everything you said, I think, summarizes the challenge already really well. I, by the way, agree. So when we do this or when companies release these statements about these computations that a quantum computer can do versus a classical computer, then I would also worry more about the real world problems versus problems specifically designed to solve.

by quantum computer efficiently solved by computer so but encryption now is one of these problems where we have a practical application we use as you said rsa for example a lot for encrypting uh communication through devices or between devices um in internet based communication network based communication so if a quantum computer is able to crack that algorithm efficiently we gotta protect against it somehow a quantum computer um they

RSA, for example, can crack RSA efficiently because the underlying structure, the underlying mathematical promise is that with a classical computer, you cannot find the prime factors of a very large numbers efficiently. Now with a quantum computer, you can do that more or less efficiently.

And that is true for other encryption algorithms too. So based on whatever mathematical promise, there may be a quantum algorithm that solves that problem efficiently just because of the features that the quantum computer has that we will not have at any time with a classical computer. So there is a class of algorithms that are summarized as post-quantum algorithms, post-quantum cryptography.

As of current understanding, and that doesn't mean that will hold true forever, as of current understanding, there is no efficient way to use a quantum computer to crack these algorithms. NIST just recently released three standards, or you can call it one standard with three algorithms, for digital signature and encryption.

And these ones are among those algorithms that, as of current knowledge, cannot be cracked easily with a quantum computer. So that's one part of the story. You will use, for encryption of file systems or communication, you will use post-quantum cryptography algorithms. But then there is more. There is other quantum technology that you can use to protect against quantum computers. So there is quantum key distribution, if you've heard of that. So quantum key distribution...

means basically I have two parties and I have a fiber optics network between these two parties and I encode my key in the quantum information say in the phase or in the polarization of photons and

and then transmit that over the fiber optics channel and through smart measurements and then exchange of information between the parties I know or the parties know if someone has listened if someone was trying to steal that key so if the parties now agree I have done quantum key distribution no one has listened we can use that key securely then they would get

the quantum the post quantum algorithm and use that key in the post quantum algorithm so i now have two components already and the third component is the secure key generation there are many random number generators out there what we supply to is a quantum random number generator so a device that uses photons or the avalanche breakdown so it's two different versions that we have

in a transistor to securely, absolutely randomly generate numbers. And these numbers, they cannot be reproduced algorithmically. So all the other algorithms that we use in the software, the random number generators, they're pseudo-random. So it's an algorithm that...

taking a seed and then generating a number. Worst case an attacker can reproduce it. So now if you bring these three together: secure key generation through a quantum device, a random number generator, secure key exchange through quantum key distribution, secure encryption through post-quantum cryptography algorithms, then you're absolutely secure.

Nice. Yeah. So that was great and concrete. And so, for example, you mentioned there NIST providing a suite of three post-quantum encryption algorithms. And so for people out there looking for those, NIST is the U.S. National Institutes of Science and Technology Institute.

famous in deep learning for being behind the MNIST. Well, so Yann LeCun in the 90s modified the NIST handwritten digit dataset, which became the MNIST dataset. It's like the hello world of deep learning problems to solve.

Cool. So yeah, so that big encryption issue seems to be something that you've now allowed our heart rates to reduce on, Florian. Another place where there's, you know, beyond encryption, there's a lot of places that quantum computing could be useful in making the world a better place. Indeed, TerraQuantum's CEO met with

the Pope in the Vatican met with Pope Francis to discuss how AI and quantum computing can be harnessed to promote human well-being, care for nature, and foster world peace in our increasingly tech-driven world. So how does TerraQuantum address its, and this is a quote from TerraQuantum themselves, TerraQuantum has a commitment to responsible innovation and the ethical implementation of quantum technology. How do you go about that practically at TerraQuantum? It's a great question.

It is just like that. So the name comes from that already. Terra is really caring about Earth, but then of course it goes beyond Earth. The idea, and I know, so especially when I say that out of the Silicon Valley, it sounds somehow flat. The idea to use technology for the good is deeply ingrained in TerraQuantum.

And for the good can mean many things. So one thing that we haven't talked about today is our medical device work, for example. So we just are about to start clinical trials for laser treatments that we do to cure osteoarthritis and to cure arteriosclerosis. That is completely novel. So what we do here is we introduce a fiber optics channel when you have arthritis between joints.

And then emit circular laser radiation, do some imaging and optimize the laser parameters using a hybrid quantum algorithm and then cure osteoarthritis that way. And the same thing can basically be done with arteries in and around your heart so we can prevent stroke and heart attack.

These are things that we would not have to do. So if we were only to focus on quantum computing and other security aspects of protecting against quantum computing, the company would still do fine. But our commitment is really to use our knowledge to develop these technologies that can help humanity. And that's why we do medical device work. We're not a medical device company, but we go through all of that, pay for the clinical trials ourselves, and then develop

try to bring it to the market, to the people, to the hospitals.

And that's true for quantum computing too. So the applications, I think many of the applications cannot be classified as good or bad in some sense, because if you think of a logistics problem, yeah, of course it's good for the nature if you have less emissions. But then if there is a problem, for example, one that we did in classifications of fatty liver, of identifying steatosis cases, that was solved using a hybrid quantum algorithm too. And we outperformed the...

We didn't have a customer for that. We just said we do it and we show it's possible and now it's out there and everyone is free to use it. So anytime we can do something like that, we just do it, even if it sometimes means we don't earn money with it. But of course, we're a company, so we sell products too.

Mathematics forms the core of data science and machine learning. And now with my Mathematical Foundations of Machine Learning course, you can get a firm grasp of that math, particularly the essential linear algebra and calculus. You can get all the lectures for free on my YouTube channel, but if you don't mind paying a typically small amount for the Udemy version, you get everything from YouTube plus fully worked solutions to exercises and an official course completion certificate.

As countless guests on the show have emphasized, to be the best data scientist you can be, you've got to know the underlying math. So check out the links to my Mathematical Foundations and Machine Learning course in the show notes or at johnkrone.com slash udemy. That's johnkrone.com slash u-d-e-m-y.

Yeah, this is interesting. And, you know, it's great that Tara has this socially beneficial commitment. For listeners that we have out there who have been listening to this episode and been thinking, wow, I too would like to make a big impact on the world with quantum computing. How can our listeners start? How can somebody become...

become a professional in, I guess, quantum computing in general, but then also quantum ML more specifically. Yeah. So if you have a background already in machine learning, then perfectly equipped already to do quantum computing, I think one of the misconceptions is that you have to be a quantum physicist or a physicist in general to do quantum computing.

You only have to be one of those if you want to build a quantum chip. If you don't, if you want to use it, so these concepts of superposition and entanglement, while initially it may be hard to wrap your mind around it, you don't have to be a physicist. The problem really is a different one. The problem is that usually when we get in touch with these concepts, we're grownups. I, when I studied, so I,

I read books early on when I was young about physics, but really how I got in touch with quantum physics was at university. So now you're 18, 19 years old, and then you learn about quantum physics, and all of a sudden, it seems everything you know about this macroscopic world is not true. So it is still true in some sense, but the fundamentals, the underlying physics is completely different.

And that is sometimes confusing. So I think if we would start to think about these concepts at an early age, maybe through quantum games, whatever it is, that would be very helpful. But now back to your question. So I think if you have some engineering background or some technical background already, you

working in data science already, then the leap to quantum computing is not that hard. It's a technology like GPUs. So people today, they don't worry about using GPUs anymore in deep learning. It's just something that you can easily add to the stack. That's what we also try to do with the software that we develop, make it easily accessible.

But of course, there is more work to do. But in the end, it can be understood, it can be learned, and it doesn't take a degree in physics. That's what I want to convey here. So is there some way, is there somewhere our listeners can go to learn about the Terra approach specifically? I mean, it sounds like a great solution to allow people who already have a machine learning background, like you said, to be able to be tackling quantum ML problems. Is there...

Is there somewhere on the Terra website that people can go to where people can get started? Maybe some open source code or some examples that allow people to get going with the Terra approach? Absolutely, yeah. We have our TQ42 platform. If you look that up, you'll find it online. We have our code documentation that will help you to get going with first examples. We have publications on the website that exactly describe how everything works.

down to the equations if that is something that someone is interested in. And then we offer TQ Academy. So TQ Academy is our own training program. So please reach out and if you're interested, we train at universities, we train companies, anyone who is interested in quantum computing. And that involves, of course, not only the Terracontam products, but the fundamentals about what is it? What is quantum computing? When does it make sense to use?

What will happen in the future? Where will it benefit AI and how, for example? Very cool. I found the TQ Academy and I will include a link to that in the show notes. Perfect. Thank you, Florian. And then for people who likewise, you know, say they go, they learn about TerraQuantum or some other quantum ML library out there. What kinds of problems do

in machine learning would you recommend as a fertile ground where there's a lot of opportunity for people to be, you know, pointing their interest, making a big impact or, you know, being able to get a lot of business? So in terms of which problems to look at, you mean? Yeah, exactly. Like, you know, are there, there's, we know that,

Quantum computing is especially well-suited to solving some kinds of problems like the traveling salesman problem or encryption. And so, yeah, in terms of machine learning, what are other kinds of application areas where there's a lot of fertile ground?

So all regression, all classification, that is fertile ground. So if you do image classification, if you do deep learning even, or large language models, then quantum computing can be interesting. Large language models, by the way, this is a very active area of research. So how can you hybridize those? We're doing a lot of work here. We're not the only ones, but this is something where still a couple of open questions need to be answered.

But generally, if you have any machine learning problem, whatever it is, there is a chance to efficiently hybridize it. In production, I would think about cameras for automated inspection. So you have a vehicle production and you want to inspect if there is some damage in the chassis when it comes out of the production line. So these are things to think about. Any classification of medical images.

really anything. In self-driving vehicles too, a lot of image classification, so that's an important thing. You may be able to train a hybrid algorithm using a quantum computer, but for the inference, you don't need it. You can run the algorithm, it's been trained, the optimal configuration has been found, but then it may not be needed to have a quantum computer for inference. So you can deploy it

in a mobile device, even if it was trained hybridized. Very cool. Thank you for those practical insights pointing us in the right direction.

Um, so going back to much earlier in the episode, you talked about huge expense being associated with people trying to have their own quantum hardware running. And a big part of this is the refrigeration. And so I'm interested in hearing your thoughts on, um, a team of researchers who earlier this year announced achieving a room temperature superconductivity in graphite, um, and,

I don't know if you know about this breakthrough or what does it mean? Or I guess even more generally, what would room temperature superconductivity mean for quantum computing? And do you think it's something that's achievable? Yes. So yes, I know of the graphite problem. That was us. Oh. Yeah.

It was our research team around our CTO here in the US, Valery Vinokur. And what they did was exactly as you said. So they achieved room temperature superconductivity at ambient pressure and ambient temperature and pressure in graphite. It's so funny that I missed it. It's now so obvious. Valery Vinokur, a condensed matter physicist who is CTO at TerraQuantum.

Yes. So that was really a remarkable piece of work that they did here. And that paves the way for applications in high temperature superconductivity. So still a lot of work needs to be done. We have to, when we think about qubits, have to think about how to take that result and design qubits around that. So that's something we're thinking about now. But then if...

you had a room temperature superconductivity, then the fridge at some point may go away. So that means I would be able to have superconducting chips potentially in mobile devices.

Very hard to imagine because then even at TerraQuantum, so thinking about what you would do with a mobile quantum chip may be sometimes challenging. We have some ideas, but I think once you have it, people will have ideas and thoughts on what to do. Right now, most of these systems are cloud-based. Some people, some companies are able to afford quantum computers and have them locally, but most of these are really in the cloud.

In terms of this high temperature superconductivity, the biggest step forward would be in quantum computing, getting rid of the fridge. And then, of course, there are many other applications. So if you just imagine you would be able to design cables that don't have any loss anymore, no resistance, you would be able to transmit energy, transmit electricity to end customers without loss completely. That would be fantastic.

But then there are many results. Remember when there was type 1 and type 2 superconductivity? By the way, Valerian team also invented or discovered type 3 superconductivity, which is now a completely novel form of superconductivity. But when type 1 and 2 were discovered in the 70s, 80s, I think around that time, then everyone would think, well, tomorrow we'll have...

all electric devices and fridges with superconducting cables. Did not happen because of other challenges. So now still some thinking needs to go into how to leverage that result, these results effectively. Gotcha, gotcha.

As I was reading this article, reskimming this article as you've been speaking about room temperature superconductivity, it also reminded me that while you are based in the Bay Area, the company, Terra, is headquartered in St. Gallen, Switzerland, which

Which is, you know, it's an interesting choice. And I'll tell you why that's particularly interesting to me. So I've been three times to the St. Gallen Symposium and I've talked about it on air many times. I think it's a great, I mean, so people, you can actually, I think for about a month more after this episode is released, if you're a graduate student anywhere in the world, you can go to symposium.org and you can write an essay there.

And based on that essay, you could be invited to the St. Gallen Symposium with all expenses paid. So your flight, your grand transportation, your food, of course, your tickets, and you get to meet amazing business and political leaders from all over the world.

So the St. Gallen is an interesting place to me having been there, you know, a number of times over the year. And I actually, I also lead the alumni. So anybody who's been to the St. Gallen symposium who lives in the U S or Canada, uh, I have them on an email list and we have events a couple of times a year, uh, mostly in New York. Um, and so, yeah, so I, I have this connection to St. Gallen. And so it's interesting to me, uh,

Why St. Gallen as Terra's headquarters? I'm not really aware of any other organizations other than St. Gallen University in that city. So I think the choice goes back to our CEO who decided this is a great place. It is a great place, has good academia around. Zurich is very close. ETH is very close. St. Gallen University is close.

We've got some collaborations with those academic institutions. Then it's co-headquartered in Germany, it's in Munich, and the same reason for that. So it's very close to innovation. It's a booming area, I would say, in technology. St. Gallen per se, in my view, not so much, but it's very well embedded into this whole academic area or surrounded by great institutions.

So therefore we find much talent in these institutions too. The same is true for Munich, it's a great place too.

That's the only reasons for why we're there. And we expand it. So, I mean, I'm calling in from San Francisco. We have a nice office here too and other nice places. Nice, yeah. I mean, those are beautiful. St. Gallen, Munich, they're beautiful parts of the world. San Francisco has lots of beauty around it. You're a short drive to lots of really nice natural beauty as well. Yes, absolutely.

Nice. Yeah, that makes a lot of sense. All right. So one last final technical kind of question for you before I get to my final questions, which is, as you look to the future of quantum technologies, what do you think is the future of quantum technologies?

what are some of the most exciting developments or applications that you foresee becoming feasible that aren't yet today? Um, maybe even in fields that aren't traditionally associated with quantum computing. You know, we talk a lot about encryption today. Uh, you know, you listed some interesting, uh, machine learning examples where you said, you know, regression, classification, deep learning soon, maybe LLMs will be able to hybridize with, uh,

a quantum chip and take advantage of that in some cases. But yeah, what do you, this has been a long-winded question, but what do you see for the future long-term as exciting applications of quantum tech? - I think, so there are many. I think what gets me excited a lot is drug design. So the ability of finding personalized medicine, even if we suffer from the same disease,

the treatment both of us may benefit may be different. However, right now, today, we would most likely get the identical treatment, almost identical treatment, medical, so the same drugs that we use. And these are all based on experiments and are 20, 30 years old sometimes. But with a quantum computer, I would be able to really do personalized medicine because it's just easy to design the right drug for your body.

That gets me excited. Everything else in terms of efficiency, of course, too. When we think about fusion, for example, nuclear fusion, that requires a lot of physical processes, simulation of physics and better understanding of the physics. That's where quantum computing can help. Optimization, the beam control needs to go really fast. So that's where quantum computing can help. So all of these areas strongly benefit from quantum computing.

new technologies such as space elevators, for example, something that you cannot really efficiently build or not build at all today using the materials that we know because nothing has the tensile strength to hold a platform in geostationary orbit. So these are things that will be possible using a quantum computer.

So that's very, very exciting, I feel. Cool. All right. And, you know, I got to say, I feel a bit bad for knocking San Francisco so hard because Golden Gate Park right in San Francisco is also beautiful. If you can get that on a warm day. It's beautiful.

And so, yeah, thank you so much for taking so much time with us today, Florian. Your time is so valuable. And so we really appreciate it. Before I let you go, do you have a book recommendation for us? A difficult question. I know you told me at the beginning that you would ask me at the end. There's so many books that I would recommend, but I thought about two that I recently read. And these are biographies.

So one is The Man from the Future, which is a biography about John von Neumann. And the other one that I really found inspiring is Well, Doc, You're In, which is a biography of Freeman Dyson. And that one, the second one I find even more inspiring. If you know Freeman Dyson, what he did and how much he contributed to so many fields, starting from space exploration, cosmology, physics, fundamental physics,

It's just amazing to read about that person specifically and how nice a person he was. So these two I found very inspiring, second one even more.

Nice. I love those. You know, we don't get enough biographies recommended on this show. And those two people, I mean, those are fascinating people. You know, they're not, it's so easy to give the like, oh, Elon Musk biography or Steve Jobs biography. Those ones, von Neumann, Dyson, these are big heavyweights in discovery and invention. Incredible. Yeah. Yeah. I wish I had time to read every book recommendation because I'd love to dig into those right now.

And the final question that I always ask my guests is how they should follow you after today's episode. This was an amazing episode where I was able to learn so much. I'm sure our audience did as well. Lots of eyes opened, lots of new possibilities percolating in people's minds as a result of quantum ML capabilities. How can people follow you after this episode to get more of your thoughts?

Thank you so much, too. It was very, very exciting. And so many great questions, a good conversation I feel we had. I'd love to spend more time chatting.

Maybe offline. Maybe we can continue here. But I must say, I don't have a big online profile. I'm on LinkedIn. So that's where you can find me. And of course, through TerraQuantum. Nice. Thank you. Well, it's nice to keep it simple. It's nice to have one place to track you. Fantastic. Florian, so much for taking the time. And yeah, maybe we could check in again in a couple of years on air and see how the TerraQuantum journey is coming along. Absolutely. I would love to. Thank you so much for having me.

Nice, I loved today's episode with Dr. Florian Neukart. In it, he covered how hybrid quantum computing combines classical computers with quantum processors using quantum capabilities only where they provide clear advantages, how quantum computing is proving practical today for optimization problems like logistics and scheduling, for simulations such as physics and chemistry simulations, and for quantum machine learning, including regression and

classification and deep learning. He talked about how current quantum computers have very short operational times, nanoseconds to microseconds due to quantum decoherence requiring careful algorithm design. He filled this in on the three key elements that are emerging for quantum safe security. That's quantum proof encryption algorithms, quantum key distribution and quantum random number generation. He filled this in on recent breakthroughs in room temperature superconductivity at TerraQuantum that could eventually eliminate the need for expensive cooling systems in quantum

computers. And he talked about how no physics degree is required to work with quantum computing. Those with machine learning backgrounds can get started right away with platforms like TQ42, the Terra Quantum Academy. As always, you can get all the show notes, including the transcript for this episode, the video recording, any materials mentioned on the show, the URLs for Florian's social media profiles, as well as my own at superdatascience.com slash 851.

Thanks, of course, to everyone on the Super Data Science Podcast team, our podcast manager, Sonia Bryovich, our media editor, Mario Pombo, partnerships manager, Natalie Jaisky, researcher, Serge Massis, our writers, Dr. Zohra Karchei and Sylvia Ogweng, and our founder, Kirill Aromenko. Thanks to all of them for producing another exceptional episode for us today. For enabling that super team to create this free podcast for you, we're deeply grateful to our sponsors. You can support this show by checking out our sponsors' links below.

which are in the show notes. And if you yourself are interested in sponsoring an episode, you can get the details on how to do that by pointing your browser to johnkrone.com/podcast. Otherwise, share this episode with folks who would love to learn about quantum computing or quantum ML, review the episode on your favorite podcasting app or on YouTube,

Subscribe if you're not a subscriber. But most importantly, I just hope you'll keep on tuning in. I'm so grateful to have you listening and hope I can continue to make episodes you love for years and years to come. Until next time, keep on rocking it out there. And I'm looking forward to enjoying another round of the Super Data Science Podcast with you very soon.