# Podcast with Amir Naveh, co-founder Classiq

My guest today is Amir Naveh, co-founder and Head of Algorithms at Classiq. Amir and I talk about his vision for Classiq, new types of quantum algorithms, whether circuit optimization is important in the long run, and much more.

Listen to additional podcasts here

# THE FULL TRANSCRIPT IS BELOW

**Yuval**: Hi, Amir. Thanks for joining me today.

**Amir**: Hi, Yuval. Pleasure to be here.

**Yuval**: So who are you and what do you do?

**Amir**: So, my name is Amir Naveh, one of the Co-Founders and Head of Algorithms at Classiq Technologies, which of course the Qubit Guy podcast is a part of.

**Yuval**: Great. After doing a couple of dozen episodes with other guests, I was asked why don’t you interview someone from the company? So, here we are. So, why did you start Classiq, together with Nir and Yehuda?

**Amir**: So I think probably most of the listeners to this podcast of course, can relate to the fact that we are on the verge of a huge shift in computing when quantum computing happens. And for us, the number one question was timing. So, we were certain that we have, in our minds, the technology needed to advance these computers, but obviously to form a quantum computing software company, you are very, very reliant on the hardware. So, when we founded this company in late 2019, we felt it’s the right time on the one hand that the hardware is getting mature and that within 2, 3, 4 years will start seeing the industrial use cases and real applications and possibly the way forward to quantum advantage, but also that there is still this huge blue ocean that really allows to build the technology stack for quantum computing from the ground up and that was what really got us excited, this opportunity to go into a new technology that is really changing the face of computing and doing it from the earliest days.

**Yuval**: In terms of your vision, what problems did you set out to solve and who did you set it to solve it for?

**Amir**: What we were thinking — or still are thinking — is that the number one challenge in quantum computing is of course the hardware. Doing the hardware, making it really work with a large amount of qubits, high fidelity and low noise. That’s an incredible engineering challenge. And what we had in mind still do, is that when these computers happen, it would really be sad, if the software weren’t able to take advantage of these amazing machines. If we really have these large quantum computers that are able to solve immense tasks, and we’re still writing application or algorithms at the gate level, very non-optimized, just putting the individual pulses or gates and designing quantum circuits that way we’re not going to be able to utilize these amazing machines.

So, what we had in mind in terms of users and value is that when you want to use these computers for real world applications and actually take advantage of them, you need another layer or layers of abstraction from the gate level up to the place where you can actually define what you want these machines to do. So, that’s the task we set out to do.

**Yuval**: And who is it for what? There are many players in the ecosystem, there are end users, there are developers, there are hardware providers. Who is this platform, the Classiq platform intended for?

**Amir**: Let me try and answer that short term and long term. Short term, I think that most of the ecosystem today is still quantum information experts, PhDs, people really understand what unitary matrices are and how they behave and how the evolution of these quantum states work. And so, in the early days, I think the software is intended for them. So, many enterprise teams are forming with these experts in quantum computing and they’re solving the domain specific challenges. And of course the software is intended for them. So, they’re able to build their applications and the quantum circuits in a hugely optimized and much better way than is otherwise possible. So, that’s if I look a year or two ahead, but if I’m trying to look maybe five years ahead or a decade ahead. So, I think it won’t necessarily be quantum information experts that are programming these quantum computers. You need to understand the problem very well and you need to understand that it fits the nature of problems that can be solved with quantum computers.

So you’d have quantum chemistry problem or an optimization problem, or some other problem that quantum computers are naturally good at. But then I would hope that our platform also enables them to describe and solve these problems on a quantum computer without really understanding all the details down to the unitary and how the error correction should work and the complex things that are necessary today.

**Yuval**: You mentioned programming today at the gate level, but is that really the case? When I look at other software packages, sometimes you see, “Here’s a VQE algorithm, just put in these values. Or here’s a Traveling Salesperson Problem, put the coordinates and we’ll figure out a solution for you.” So, how true is it to say that other solutions are really at the gate level versus block or predefined function level?

**Amir**: That’s a really good question. I think that’s one of the things that we really should emphasize on. What Classiq is doing and what the problem really is today for quantum software developers? So, if you want, let’s take your example of VQE for example. So, in essence, you have this molecule and you know how to write down the Hamiltonian of the molecule. And you’re trying to find, for example, the ground state for this molecule. But then, so from this to the gate level circuit, from the actual CNOT gates and the Hadamard gates, there is a huge space of design flexibility. You can order the terms in the Hamiltonian in many different ways if you’re doing Trotterization and you can order the different individual gates in different ways, and you can do different Trotterization schemes, and you can trade off between the error or accuracy of the algorithm and the depth or the number of CNOT gates or the number of qubits you’re using.

So all of this designs space is something that in essence, what you want to do is solve the ground state of the molecule. But from there to the quantum circuit, there are many, many choices you need to make. And when you’re making these choices manually, then you’re losing out on huge amounts of optimization. So, it’s orders of magnitude non-optimized, and also you are working on things that really aren’t what you should be working on. These are things that should be automated and done with a computer or with a platform that solves these technology problems. And same thing I could say for optimization problems or for search problems or for any other problem that really, it’s every time it’s a different problem, right? You’re solving a different molecule, a different optimization problem. So, it’s not just plug in this prebuilt circuit that we have for you or this quantum for your transform or whatever rigid building block you have.

**Yuval**: But let’s dive a little bit deeper into that if I can. So, I understand that if I’m a quantum information scientist and I want to run VQE algorithm, then it would be very difficult for me to manually create a VQE circuit that’s optimized. But if I’m a company that specializes in VQE circuits, then maybe I’ve work months or man years, and created this wonderful VQE circuit, is that optimized enough or does it need to be re-optimized for every single problem? Why would a single circuit not just work in most cases, work good enough in most cases?

**Amir**: Let me give you two examples. So, one example is, let’s say you are running this circuit on whatever hardware. I don’t know if you designed it for a specific hardware in the first place, but let’s say you’re working on a specific machine, and now you want to move to another machine. And let’s say these two machines have a different level of fidelity for their two-qubit gates. So, the design choices you made, even if you really did the hard work and optimized, and these are things that aren’t possible to optimize manually. So, maybe you built for yourself this optimization, custom made for your very specific circuit. And now you’re moving it to another hardware with different parameters for the gate fidelity. So, you’re going to have to redefine all of your optimization problem, because the choices you made were optimized only for this one specific hardware. Same thing for the hardware connectivity.

Same thing for, okay, so you’ve solved one specific problem for one specific molecule. Now you’re moving to a slightly different molecule. So, is it the same or is it different, but even if you were working with one specific hardware, with one specific gate, with one specific error and everything is fully, fully, fully predetermined, still, the problem is very, very hard to solve. So you’re going to have to kind of internally develop the technologies that we’re working on. And, these are hard problems. These are things that one or two people working independently, won’t be able to solve. You need a company like ours with dozens of engineers that are solving these optimization problems. So, even if it’s a very specific case, you won’t be able to really compete without the technologies.

**Yuval**: How important is optimization in the long run? What I mean by that is if I have a 50-qubit computer that I understand, I need to optimize to squeeze every ounce of performance from that computer. But maybe in a couple of years, I have a 5,000-qubit computer. And so, maybe it doesn’t matter if my circuit takes 2000 qubits or 2,700 qubits. How important is optimization in the long run?

**Amir**: The first thing to keep in mind is that when we’re saying optimization, we’re not talking about twofold optimization. We’re talking about a thousandfold optimization. So, we can go down from 100,000 qubits, which is what you would need if you’re not doing any auxiliary management or any reuse of qubits and just designing naively to thousands of qubits. So, it’s a hundred fold or a thousand fold on the optimization on the number of qubits, the circuit depth, the circuit accuracy. So, of course, that becomes significant. But I would say, okay, so if you have larger machines with better accuracies and fidelity, that’s great. So, you can go ahead and run even more complex problems. You can expand the envelope of what you’re allowing your computer to do.

So, just like when you have computers today and you’re always searching, 10 years ago, one gigabyte of RAM sounded incredible. Now, you have on a standard laptop, 16 or 32 gigabytes of RAM and you need that. All the applications are designed for that. And you could say, okay, so 1 billion transistors on a chip is probably enough, but still you’re trying to go to 5 billion and to 40 billion. So, exactly the same sense, I don’t think that, 2000 qubits or 10,000 qubits will be enough. You’re always going to want to run state of the art applications on the computers. So, I’d say optimization is of the highest importance.

**Yuval**: Today there are a dozen, maybe less than a dozen important quantum algorithms, right? There’s VQE and QAOA and Grover and Shor’s of course and HHL and so on, but they’re not hundreds of different algorithms there, a few. On the other hand, when you look at classical programming, I was looking it up the other day, I think that Wikipedia lists over 40 different algorithms that you can use for sort. Why does this gap exist in your opinion? And does the Classiq platform help create new algorithms for the future?

**Amir**: So I’d say, I look at it a bit differently. Let’s look at one of the short list you mentioned of the QAOA algorithm. So, that’s a framework for solving, I mean, in our platform, you have implemented Traveling Salesperson, Knapsack, Integer Optimization, Max Vertex Cover. So, basically any, any optimization problem, any discrete optimization problem you can think of, can be mapped into this framework. So, it’s not, you can think of it not as one algorithm, but as hundreds or thousands of different optimization problems that are probably important to any computing intensive industry, right from logistics to supply chain to financial optimization. And the same thing I can say for VQE, which solves all problems in quantum chemistry in essence, and the same I can say, of course, for Grover, which I mean search problems, you can map 3-SAT into search problems, and you can map Bitcoin mining into search problems. So, it’s really, if I look at the applications or useful algorithms, I can map thousands of problems into these, even these three or four algorithms.

And also I’m hoping that, we’re only at the beginning. So, the number of people working on these quantum algorithms and applications is growing year by year. And these are a few frameworks, but I think things that are maybe not as fundamental as Shor’s algorithm, but frameworks for solving real world problems, I think they will evolve and expand, every year we’ll get more and more of these.

**Yuval**: If you think about Classiq three years from now, or five years from now, what do you see? What do you see the platform doing that it’s not doing today? How do you see people using it? What should we be expecting?

**Amir**: So allow me as a founder to dream big, right and of course, we’ll all have to see what the future has in store for us, both in terms of the hardware and for our specific company. But if I try and give a vision to where we want to be in five years, so quantum software really is at the beginning. If the people listening have had any experience with really trying to design quantum software, quantum algorithms, you know, it’s at a whole, whole different level than what’s happening in classical software. Maybe like five or six decades ago, we were in classical software in the same layers. So, Classiq is building these layers of abstraction. And in five years from now, if all goes well, then we will be the platform that enables you to take your problems, whatever industry you’re in, pharmaceutical, financial, optimization in whatever industry, aerospace, defense, any, any, really any industry that uses computers, which is hard to find one that doesn’t, and you map your problems with our platform into optimized quantum circuits, and then run them on the hardware that sits on the cloud.

So, right now, we’re also integrated with Azure Quantum and AWS braket and IBM machines and several others. So, this I think will stay like that. And then you can, you have the whole workflow in your hands doing the optimization and the execution of these algorithms and the power is yours to use these incredible machines. We hope to be the leading quantum software company and users from all industries will use this platform to really solve their problems. I hope we’ll be there in five years.

**Yuval**: Many algorithms today or many use cases today, involve the combination of classical programming and quantum programming. Because even if you use a quantum computer, you need to get the data from someplace. You need to visualize it sometimes in some iterative algorithms, you are essentially changing the quantum circuit as you run. So, do you see classical programming environments merging with quantum programming environments? Or do you see them as two separate disciplines?

**Amir**: For sure, they’ll have to talk together. This is something actually we’re quite focused on, I’d say in the short to midterm in our R&D focus. How to really take these hybrid schemes, whether it be VQE or QAOA where you have these iterative schemes or sometimes doing things like amplitude destination or amplitude amplification, which requires some classical post-processing measurements within the circuits and how to handle them. So, it has to be combined, but I think that’s in my view, that’s a challenge, but not the most difficult challenge because in the end, what you need to do, is do the right APIs and have the things talk with one another in the right interfaces. But it’s not a fundamental technology problem. It’s more of a creating standardization and it will happen as the industry matures. I’m not very worried about that, but I think it’s a very, it’s a focus that we all need to have in mind.

**Yuval**: As we get close to the end of our conversation today, I wanted to ask you about other software problems in quantum computing. So, there’s certainly the problem of how you create circuits and how you translate an algorithm or a recipe for algorithm into a working circuit, but are there other problems for instance, if the number of qubits gets larger, how do you simulate? How do you debug a quantum problem? Is that an issue and is Classiq thinking about doing something about it?

**Amir**: Yes. Simulation of course becomes impossible after maybe 40 or 50 Qubits, but like you’re saying to analyze, to debug, to validate your answers, of course, these are critical problems. I can, I can state a few more I think that, and the happy part is that all of these problems are also true in the classical world. I mean, it’s very, very hard to validate the logic of a large chip. These are hard computational problems that huge teams are working on in the classical world. Same for visualizing a billion chip transistors. So, when you have 10,000 qubits, of course, it’s going to need to visualize it in a different way than how we’re viewing circuits today. So, yeah, if you look at how we treat visualizing circuits at the functional level, I think we’re ahead of what is done outside of Classiq, but I think these are all challenges that will have to be addressed. And as the industry goes on, these things will be, it’s hard challenges. Each of these words requires its own company and a lot of development power.

**Yuval**: One question that I get a lot is if you have a family friend or a relative, that’s just going into college and is interested in quantum computing, what should they study? Is it physics? Is it math? Is it English lit? What should they be focused on if they want to get it into quantum computing?

**Amir**: The school solution today is to learn quantum information, do a PhD, either in theoretical computer science or in physics and dive into these incredible domains. And, of course there are many really interesting problems to solve there. And, that’s the best, that’s the most straightforward way. But I would also say that coming from a development background, if you’re a very strong software developer, then the barrier is not that large to understand how this new paradigm of computing works and adapt yourself to that. So, I think even experienced software engineers without the quantum experience, and some of the people in our company have done that, can make the transition to, I’d call them quantum software engineers. So, that’s another way I think, to kind of get into this domain in, definitely in the future.

**Yuval**: Excellent. And, we’re definitely hiring as far as I know. Amir, how can people get in touch with you to learn more about you and your work?

**Amir**: So, best way is to send me an email at amir@classiq.io, also message me on LinkedIn or I think those are the best ways to reach me. I try and be available.

**Yuval**: Excellent. Thank you so much for joining me today.

**Amir**: Thank you Yuval for this episode and for others as well. I’m a very enthusiastic listener and I enjoy it very much. Thank you.

# About “The Qubit Guy’s Podcast”

Hosted by The Qubit Guy (Yuval Boger, our Chief Marketing Officer), the podcast hosts thought leaders in quantum computing to discuss business and technical questions that impact the quantum computing ecosystem. Our guests provide interesting insights about quantum computer software and algorithm, quantum computer hardware, key applications for quantum computing, market studies of the quantum industry and more.

If you would like to suggest a guest for the podcast, please contact us.

*Originally published at **https://www.classiq.io**.*