Podcasts

Podcast with David Von Dollen, Volkswagen America

12
January
,
2022

David von Dollen is a lead data scientist for Volkswagen America. David describes Volkswagen’s experience with quantum computing. He reveals lessons learned over a period of five years, during which Volkswagen prototyped, tested and deployed several quantum solutions.

Listen to additional podcasts here

THE FULL TRANSCRIPT IS BELOW

Yuval: Hello, David. Thanks for joining me today.

David: Hi. How's it going, Yuval? Glad to be here.

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

David: My name is David Von Dollen, and I've been with the Volkswagen Group for the past six years. I was a founding member of the Volkswagen Group quantum computing team that is based out of the Data:Lab in Munich, Germany, and I also work for Volkswagen Group of America as part of their AI team, working on research and applications in AI and high-performance computing.

Yuval: That's great. So Volkswagen, I think, is one of the great test cases for using quantum computing in an industrial way. Tell me about the project. When did it get started? Why did it get started? How was it started in the organization? I'm really curious. Anything you can tell me would be great.

David: Sure. So back in 2016, I was working out of the Volkswagen Code Office based in San Francisco, and my colleague, Florian Neukart, came over from Germany to work in the U.S. for a while and work on innovation topics. And as the Code Office was an Innovation Office, we were tasked with researching new technologies and applications in regards to the value chain for Volkswagen and the mobility space. We decided to choose quantum computing as a topic area to investigate into. As we kind of found that the technology was emerging and that in reviewing the history of the literature, it had some enormous potential to unlock value in the mobility space, as well as the logistics and materials design and optimization. And so that kicked off a whole effort into the diverse kind of set of research that our team is undertaking.

But the first project that we worked on was traffic flow optimization with a quantum annealer, essentially, where we looked at taxi routes over the city of Beijing and wanted to find a global configuration of routes for taxis that lowered the overall energy or flow or optimized the flow, maximized the flow, through the road network. And so that project kicked off kind of the quantum computing effort for Volkswagen. And since then, in 2019, we launched our first production application, leveraging the work that was done for that project, for optimization of bus routes for the Web Summit in Lisbon. Optimizing routes chosen by buses from the airport into the downtown area. And yeah, the team continues to investigate different topic areas, such as quantum machine learning. Our team coauthored a paper with Google in 2020 on the TensorFlow Quantum library. And we collaborated with them on that effort. And we are also looking into materials design and optimization, as I mentioned, as well as other optimization applications.

Yuval: And I think there was also the paint shop project?

David: Yep. Our team first investigated a binary paint shop encoding, leveraging QAOA on gate model computers. And we generalized that to multi-car optimization for the paint shop using a quantum annealer. And so that's another application that is currently being pushed into production.

Yuval: It's fascinating that you say that you started in 2016. What was the impetus? Was it top down or bottom up? Was it the CIO saying, "Oh, we should really look at quantum, and hey Florian, or this person or that person, look into that," or did it come from an engineer or a researcher that tried to push it up the chain?

David: I think it was a combination of both. We were lucky to have some investment from top management. Martin Hofmann was Global CIO at the time. And Abdallah Shanti, who is a CIO for Volkswagen Group. They put a lot of time and effort into allowing us the bandwidth to look into new technologies and investigate innovation to bring value back to the group. So there was some support there. I think a lot of it was also kind of self-driven from the bottom up. It was up to us to kind of find the direction and prove out the value and figure out, "How can we apply these new technologies in ways that are new, novel, and bring value to the company?"

Yuval: If you could do everything again, what would you do differently? What were the things you did right? What were the things you did wrong? And specifically, if I may generalize it for people who are listening to us and are a little less advanced than Volkswagen is, what do you think they should pay attention to as they try to roll quantum into their enterprise?

David: That's a great question. So I think along with other technologies, such as AI, one of the main challenges is figuring out how the business problem can translate down to a technical level. And I think we were lucky in the sense that we had backing by top managers within the company. And I think in order for these projects to be successful, there has to be a real synergy between business and IT teams to really understand the business problem that's being solved, because at the end of the day, the problem is going to determine whether or not a quantum computer is applicable, as well as the value that you can create with the quantum computer. So I would say that is one of the main challenges that I think... You can build up this incredible competency in quantum computing, but at the end of the day, you have to be able to create those horizontal relationships with business and put that in the hands of a business user or put that into production to be able to translate value.

And I think if companies can figure out how to support that collaboration and prioritize that, then they'll be a couple steps ahead in the game.

Yuval: So just returning to the paint shop project, but we can take any other project. Were you able to communicate with the supervisor of the paint shop and say, please explain to us what your problems are and what would constitute success? And, were the paint shop people involved in the daily or weekly progression of the project, or was it, "Here's the problem and we'll come back in six months and show you a solution?"

David: So we're lucky in that Volkswagen is a very large company with a lot of different components across its value chain. And so there's a lot of different opportunities to optimize and apply this new technology. We were lucky to be able to iterate with our partners in the paint shop to really, as you said, define those success criteria, refine the problem, refine the business requirements, and iterate quickly to get to our solution. Does that answer your question?

Yuval: It does. I mean, so they were involved not just in kicking off the project but during the project itself, I understand.

David: Yeah. I mean in general we apply an agile methodology where we set up requirements or tests or hypotheses ahead of time that we want to investigate. And then using an agile sprint cycle iterative loop, we can construct feedback with the business to say, "Okay, hey, in the last sprint cycle, we had this finding. We were able to sample solutions in this energy spectrum using the quantum annealer," for example, and assess whether or not that meets our sprint goal or our success criteria. And then the future sprint, we continue to iterate and refine that result.

Yuval: Let's talk about the composition of the ideal quantum team. And again, looking at companies who are perhaps building this and just want to learn from your experiences. I would guess that there needs to be people who understand quantum, maybe physicists; probably computer science people; I believe you have a PhD in computer science. You spoke about at least someone that can translate the business problem or understands the business problem. What else, if anything, is missing from a really good quantum team for an enterprise?

David: So what's interesting is that we have so many different emerging hardware technologies right now in the quantum space. And as well, we're in the NISQ era. So, for example, if we're looking at quantum annealers, actually on a high level you can frame an optimization problem as a QUBO or an Ising model. But then you can also optimize parameters on the chip, such as the chain strength or the couplings. And so I think that having an understanding on the hardware level is good. So you want to have somebody who has an understanding of all the hardware technologies and can understand the strengths and limitations. For example, different types of QPUs and the embeddings of gate model. How long are the coherence times? What's the maximum qubit values that we can use? Those types of things, that will allow the team to kind of understand the strengths and weaknesses of different hardware.

I think having software engineers on the team are always a good fit because as you're developing solutions, you may need to develop a UI or figure out how to push the quantum service into production, and how does that look in regards to security or authentication and things like that. In regards to research, the way we kind of have it set up now is we have experts within different areas. So we have people who are focusing on optimization; quantum machine learning; and there's overlap between all the spaces, but other people are looking at quantum chemistry and material simulation. So I think, depending on your problem area that you want to focus on, building up experts in those areas, as well as some generalists that you can train up and can grow within the team.

And then, yeah. So I think if you can develop that, as well as maybe a business facing role, business analyst or team lead, that can help the business understand how to augment their process and derive value with the technology. I could see all those roles being a really strong team.

Yuval: Once you had the prototype working, once you had data and you could show the internal customer in this case, "Hey, this is working. This is great. It can save you time." Was it difficult to move it into production? Were there issues of uptime or connectivity to existing enterprise systems? Was it more difficult or was it easier than you initially envisioned that it would be to move it from experimentation and prototyping into production?

David: I do think that's a central challenge and one that we're still trying to tackle in other spaces, such as AI. How do we get a team that's primarily focused on research and development to push something into production? Should that team own those production deployments? Should there be another team, right? Those are still kind of open questions I think organizations can answer on an individual basis. But I'll point to the Lisbon 2019 shuttle example. So we originally implemented a prototype and then we ended up using a D-Wave API service that we would call routinely every 12 seconds to optimize the routes. And while we were up, I don't think we had any service interruptions or downtime. But I think if you're going to put stuff in production, having a good testing strategy in place and figuring out all the architecture considerations and things like authentication are definitely good things to have in place.

Yuval: What can the industry do? I mean, you are in Volkswagen and you're using certain quantum computers, but if you look at the broader ecosystem, what can we do to make your life easier? And companies like Volkswagen, to make their lives easier to move into quantum?

David: So I think there's a couple things. The first thing I think that I'm seeing is, since we have so many different hardware platforms at the moment, developing a good middleware that can allow a researcher or engineer the ability to design, test, and evaluate different algorithms quickly while switching out backends. I could see that as a really powerful solution.

The other thing I think comes in the form of, and I think this might have been touched on in other podcasts, but in the form of training, I think as we've gone through the last five years, it's gotten easier and easier for people to get into quantum computing. And there's been more and more resources available for people to learn more about the canonical algorithms, how to compose circuits from gates, what are the strengths and limitations currently with the hardware? So I think giving people more of a chance to get trained up is another area of growth that I could see. And I think those are the two things that... As well as helping businesses realize how to extract value with quantum computers. I could see those points as all being beneficial and helpful.

Yuval: And as we get closer to the end of our discussion today, I was interested in your predictions for next year. What do you think is going to happen in the quantum world that you are potentially excited about?

David: That's a really great question. So I think in 2022, we're going to continue to see innovations in regards to quantum machine learning, and people answering kind of the big questions around, "What's the representational capability of a quantum neural network?" for example; or we're going to see new applications of quantum computing to business across different business domains. Whether it's combinatorial optimization or materials discovery and design. I think we're going to continue to see new hardware and software platforms emerge. And I know at the danger... I know we're kind of entering a peak of a hype cycle. So while I fear the drop, but I'm optimistic that in the short to medium term, that we will be able to find use cases and value from quantum-inspired solutions or quantum-enhanced solutions, while we get to a point where we get beyond the NISQ era and have millions of fault-tolerant qubits, and can do things with that level of quantum computation.

But to me, if you find a quantum-inspired algorithm that performs better than classical, that increases value for your company. I mean, to me, that seems to be a win. So I'm hopeful that as we move through this hype cycle, we create lasting things that we can learn from and technological improvements, and bring up a generation of scientists that will eventually contribute to solutions that we may not even know exist today.

Yuval: Absolutely. David, how can people get in touch with you to learn more about your work?

David: Well, I could give out my work email, which would be David.VonDollen@audi.com.

Yuval: Excellent. Well, thank you so much for joining me today.

David: Thank you, Yuval. It's been a pleasure.



David von Dollen is a lead data scientist for Volkswagen America. David describes Volkswagen’s experience with quantum computing. He reveals lessons learned over a period of five years, during which Volkswagen prototyped, tested and deployed several quantum solutions.

Listen to additional podcasts here

THE FULL TRANSCRIPT IS BELOW

Yuval: Hello, David. Thanks for joining me today.

David: Hi. How's it going, Yuval? Glad to be here.

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

David: My name is David Von Dollen, and I've been with the Volkswagen Group for the past six years. I was a founding member of the Volkswagen Group quantum computing team that is based out of the Data:Lab in Munich, Germany, and I also work for Volkswagen Group of America as part of their AI team, working on research and applications in AI and high-performance computing.

Yuval: That's great. So Volkswagen, I think, is one of the great test cases for using quantum computing in an industrial way. Tell me about the project. When did it get started? Why did it get started? How was it started in the organization? I'm really curious. Anything you can tell me would be great.

David: Sure. So back in 2016, I was working out of the Volkswagen Code Office based in San Francisco, and my colleague, Florian Neukart, came over from Germany to work in the U.S. for a while and work on innovation topics. And as the Code Office was an Innovation Office, we were tasked with researching new technologies and applications in regards to the value chain for Volkswagen and the mobility space. We decided to choose quantum computing as a topic area to investigate into. As we kind of found that the technology was emerging and that in reviewing the history of the literature, it had some enormous potential to unlock value in the mobility space, as well as the logistics and materials design and optimization. And so that kicked off a whole effort into the diverse kind of set of research that our team is undertaking.

But the first project that we worked on was traffic flow optimization with a quantum annealer, essentially, where we looked at taxi routes over the city of Beijing and wanted to find a global configuration of routes for taxis that lowered the overall energy or flow or optimized the flow, maximized the flow, through the road network. And so that project kicked off kind of the quantum computing effort for Volkswagen. And since then, in 2019, we launched our first production application, leveraging the work that was done for that project, for optimization of bus routes for the Web Summit in Lisbon. Optimizing routes chosen by buses from the airport into the downtown area. And yeah, the team continues to investigate different topic areas, such as quantum machine learning. Our team coauthored a paper with Google in 2020 on the TensorFlow Quantum library. And we collaborated with them on that effort. And we are also looking into materials design and optimization, as I mentioned, as well as other optimization applications.

Yuval: And I think there was also the paint shop project?

David: Yep. Our team first investigated a binary paint shop encoding, leveraging QAOA on gate model computers. And we generalized that to multi-car optimization for the paint shop using a quantum annealer. And so that's another application that is currently being pushed into production.

Yuval: It's fascinating that you say that you started in 2016. What was the impetus? Was it top down or bottom up? Was it the CIO saying, "Oh, we should really look at quantum, and hey Florian, or this person or that person, look into that," or did it come from an engineer or a researcher that tried to push it up the chain?

David: I think it was a combination of both. We were lucky to have some investment from top management. Martin Hofmann was Global CIO at the time. And Abdallah Shanti, who is a CIO for Volkswagen Group. They put a lot of time and effort into allowing us the bandwidth to look into new technologies and investigate innovation to bring value back to the group. So there was some support there. I think a lot of it was also kind of self-driven from the bottom up. It was up to us to kind of find the direction and prove out the value and figure out, "How can we apply these new technologies in ways that are new, novel, and bring value to the company?"

Yuval: If you could do everything again, what would you do differently? What were the things you did right? What were the things you did wrong? And specifically, if I may generalize it for people who are listening to us and are a little less advanced than Volkswagen is, what do you think they should pay attention to as they try to roll quantum into their enterprise?

David: That's a great question. So I think along with other technologies, such as AI, one of the main challenges is figuring out how the business problem can translate down to a technical level. And I think we were lucky in the sense that we had backing by top managers within the company. And I think in order for these projects to be successful, there has to be a real synergy between business and IT teams to really understand the business problem that's being solved, because at the end of the day, the problem is going to determine whether or not a quantum computer is applicable, as well as the value that you can create with the quantum computer. So I would say that is one of the main challenges that I think... You can build up this incredible competency in quantum computing, but at the end of the day, you have to be able to create those horizontal relationships with business and put that in the hands of a business user or put that into production to be able to translate value.

And I think if companies can figure out how to support that collaboration and prioritize that, then they'll be a couple steps ahead in the game.

Yuval: So just returning to the paint shop project, but we can take any other project. Were you able to communicate with the supervisor of the paint shop and say, please explain to us what your problems are and what would constitute success? And, were the paint shop people involved in the daily or weekly progression of the project, or was it, "Here's the problem and we'll come back in six months and show you a solution?"

David: So we're lucky in that Volkswagen is a very large company with a lot of different components across its value chain. And so there's a lot of different opportunities to optimize and apply this new technology. We were lucky to be able to iterate with our partners in the paint shop to really, as you said, define those success criteria, refine the problem, refine the business requirements, and iterate quickly to get to our solution. Does that answer your question?

Yuval: It does. I mean, so they were involved not just in kicking off the project but during the project itself, I understand.

David: Yeah. I mean in general we apply an agile methodology where we set up requirements or tests or hypotheses ahead of time that we want to investigate. And then using an agile sprint cycle iterative loop, we can construct feedback with the business to say, "Okay, hey, in the last sprint cycle, we had this finding. We were able to sample solutions in this energy spectrum using the quantum annealer," for example, and assess whether or not that meets our sprint goal or our success criteria. And then the future sprint, we continue to iterate and refine that result.

Yuval: Let's talk about the composition of the ideal quantum team. And again, looking at companies who are perhaps building this and just want to learn from your experiences. I would guess that there needs to be people who understand quantum, maybe physicists; probably computer science people; I believe you have a PhD in computer science. You spoke about at least someone that can translate the business problem or understands the business problem. What else, if anything, is missing from a really good quantum team for an enterprise?

David: So what's interesting is that we have so many different emerging hardware technologies right now in the quantum space. And as well, we're in the NISQ era. So, for example, if we're looking at quantum annealers, actually on a high level you can frame an optimization problem as a QUBO or an Ising model. But then you can also optimize parameters on the chip, such as the chain strength or the couplings. And so I think that having an understanding on the hardware level is good. So you want to have somebody who has an understanding of all the hardware technologies and can understand the strengths and limitations. For example, different types of QPUs and the embeddings of gate model. How long are the coherence times? What's the maximum qubit values that we can use? Those types of things, that will allow the team to kind of understand the strengths and weaknesses of different hardware.

I think having software engineers on the team are always a good fit because as you're developing solutions, you may need to develop a UI or figure out how to push the quantum service into production, and how does that look in regards to security or authentication and things like that. In regards to research, the way we kind of have it set up now is we have experts within different areas. So we have people who are focusing on optimization; quantum machine learning; and there's overlap between all the spaces, but other people are looking at quantum chemistry and material simulation. So I think, depending on your problem area that you want to focus on, building up experts in those areas, as well as some generalists that you can train up and can grow within the team.

And then, yeah. So I think if you can develop that, as well as maybe a business facing role, business analyst or team lead, that can help the business understand how to augment their process and derive value with the technology. I could see all those roles being a really strong team.

Yuval: Once you had the prototype working, once you had data and you could show the internal customer in this case, "Hey, this is working. This is great. It can save you time." Was it difficult to move it into production? Were there issues of uptime or connectivity to existing enterprise systems? Was it more difficult or was it easier than you initially envisioned that it would be to move it from experimentation and prototyping into production?

David: I do think that's a central challenge and one that we're still trying to tackle in other spaces, such as AI. How do we get a team that's primarily focused on research and development to push something into production? Should that team own those production deployments? Should there be another team, right? Those are still kind of open questions I think organizations can answer on an individual basis. But I'll point to the Lisbon 2019 shuttle example. So we originally implemented a prototype and then we ended up using a D-Wave API service that we would call routinely every 12 seconds to optimize the routes. And while we were up, I don't think we had any service interruptions or downtime. But I think if you're going to put stuff in production, having a good testing strategy in place and figuring out all the architecture considerations and things like authentication are definitely good things to have in place.

Yuval: What can the industry do? I mean, you are in Volkswagen and you're using certain quantum computers, but if you look at the broader ecosystem, what can we do to make your life easier? And companies like Volkswagen, to make their lives easier to move into quantum?

David: So I think there's a couple things. The first thing I think that I'm seeing is, since we have so many different hardware platforms at the moment, developing a good middleware that can allow a researcher or engineer the ability to design, test, and evaluate different algorithms quickly while switching out backends. I could see that as a really powerful solution.

The other thing I think comes in the form of, and I think this might have been touched on in other podcasts, but in the form of training, I think as we've gone through the last five years, it's gotten easier and easier for people to get into quantum computing. And there's been more and more resources available for people to learn more about the canonical algorithms, how to compose circuits from gates, what are the strengths and limitations currently with the hardware? So I think giving people more of a chance to get trained up is another area of growth that I could see. And I think those are the two things that... As well as helping businesses realize how to extract value with quantum computers. I could see those points as all being beneficial and helpful.

Yuval: And as we get closer to the end of our discussion today, I was interested in your predictions for next year. What do you think is going to happen in the quantum world that you are potentially excited about?

David: That's a really great question. So I think in 2022, we're going to continue to see innovations in regards to quantum machine learning, and people answering kind of the big questions around, "What's the representational capability of a quantum neural network?" for example; or we're going to see new applications of quantum computing to business across different business domains. Whether it's combinatorial optimization or materials discovery and design. I think we're going to continue to see new hardware and software platforms emerge. And I know at the danger... I know we're kind of entering a peak of a hype cycle. So while I fear the drop, but I'm optimistic that in the short to medium term, that we will be able to find use cases and value from quantum-inspired solutions or quantum-enhanced solutions, while we get to a point where we get beyond the NISQ era and have millions of fault-tolerant qubits, and can do things with that level of quantum computation.

But to me, if you find a quantum-inspired algorithm that performs better than classical, that increases value for your company. I mean, to me, that seems to be a win. So I'm hopeful that as we move through this hype cycle, we create lasting things that we can learn from and technological improvements, and bring up a generation of scientists that will eventually contribute to solutions that we may not even know exist today.

Yuval: Absolutely. David, how can people get in touch with you to learn more about your work?

David: Well, I could give out my work email, which would be David.VonDollen@audi.com.

Yuval: Excellent. Well, thank you so much for joining me today.

David: Thank you, Yuval. It's been a pleasure.



See Also

No items found.

Start Creating Quantum Software Without Limits

contact us