ポッドキャスト

Podcast with Tom Marshall, Bloomberg

26
January
,
2022

My guest today is Tom Marshall, software engineer from Bloomberg. Tom and I spoke about deploying quantum computing in data centers, his assessment of various quantum modalities, killer applications for financial services and much more.

Listen to additional podcasts here

THE FULL TRANSCRIPT IS BELOW

Yuval: Hello Tom, and thanks for joining me today.

Tom: Happy to be here.

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

Tom: My name is Tom Marshall, I'm an experimental physicist. I'm doing software architecture and security for Bloomberg, and I've been paying a lot of attention to quantum computing.  And all of my opinions are my own.  I'm not speaking for Bloomberg here in any way.

Yuval: So, are you more on the software side, or more on the physics side, or some combination of both?

Tom: Well, my day job is purely software, but, and again, I'm a physicist by education and nature, so you can't take that out of me.

Yuval: As am I. I have a master's in physics, but I don't do physics on a daily basis. Let me ask you a software question. We see companies doing proof of concepts, and some of them are more successful than others. And some people are saying, "Okay, what would it take to move these proof of concepts from the sandbox to a production environment?" How do you see quantum fit in enterprise computing architecture? I mean, would the IT folks manage it just like they manage cloud resources? What do you see as the issues of integrating quantum into enterprise computing architecture?

Tom: It's a good question. I think it's a lot more complicated than that. First of all, I always like to say, there is no cloud; it's just somebody else's computer. So, where the physical hardware is located is not too terribly relevant to typical commercial end-users. But the issue, as I see it, is from the moment something can be done that is of commercial interest to someone, until the time that it is actually a commodity and we can say, how do we want to deploy it? And should AWS host it, or should we buy our own? Or whatever. I think that there will be a period of scarce resources and quite a bit of disruption in industry, and different industries will react differently. And again, the financial services industry, I think, will be especially volatile. So I think it's going to be a mess, and the physical deployment is the least of anybody's concerns until it becomes a commodity, and then the data center people will figure out how to manage it.

Yuval: If the physical location is not the issue, what is the issue? Is it integrating it into data input? Is it putting it in a workflow? Is it a service level management issue? What is the issue?

Tom: Scale. I think that, for instance, if PsiQuantum is successful, they have a technology that will scale relatively quickly. I am not as comfortable that the transmon superconductor machines will scale as friendly as you like. You can build a lot of them and they'll work, but they won't be 24/7 commodity reliable, I think. They will be expensive to maintain, and therefore, I think they'll be a scarce resource for times frame of years until they get it worked out because it'll do something useful, it will be very valuable, but you won't have a million of them in a data center. That's what I see. I'd like to be wrong on that because I'm not in favor of disruption, but I see disruption in the future.

Yuval: So if you want to run portfolio optimization, for instance, just to say, common financial services use case, how often do you run it? How, how many times a day, how many times a second do you run it?

Tom: I do know that it's compute limited. I do know that, could it be done in real time, that's a service that could be offered to day traders going to market with a balanced risk portfolio thing in and out in real time. I mean, the poster child is derivatives pricing, very slow, very expensive. If you could just price the basket of derivatives and in and out, that would, I think, revolutionize the market. Again, I'm not a financial guy, but it seems that we're so compute limited that that's where the disruption will be.

Yuval: I agree with you. The point I was trying to make, or trying to ask is: some computations need to happen in real time or constantly through the day, and others could happen in batches. Maybe it's not a financial services question. Maybe it's a shipping company that needs to recalculate the routes once a day or a couple of times a day. And so perhaps the 24/7 availability of compute resources is not as critical for some applications.

Tom: Oh, I completely agree, and those are somewhat less disruptive. I don't know how Amazon handles logistics, but let's just say that Amazon could greatly improve their logistics. Okay, fine. So my socks are two pennies cheaper, or Bezos gets two pennies richer, but the world doesn't change because you're doing batch computations there. So I hear what you're saying. I simply haven't focused on those applications. They're probably good, I mean, they're important, but that's not where I see the biggest social impact and economic impact on quantum computing.

And again, something that we've talked about, if you could say the so-called Greeks, the partial derivatives of risk and value, if you could calculate them in real time would be great, but if you cached them once a day and then have classical algorithms hit that cache, that's great, that's wonderful. Again, it will be opportunistic, it will not be that some product manager specifies what they want, it'll be, well I can give you this. Oh, good, I'll run that once a day, and then get the cache. Or, oh I can do this in real time because I can. So anyway, that's the sense of how I'm reacting to your comments.

Yuval: I understand, but it seems that there are many potential use cases in financial services, some more disruptive than others. And I think that there's a consensus amongst financial services companies that quantum holds opportunities or potential opportunities for true breakthroughs. In that context, we see many financial services firm publish results of research in academic papers. Is that just a recruiting strategy? Do you expect that to continue into the future, if indeed you have some true competitive advantages in quantum?

Tom: Well again, this is my cynical view of the world, but right now the most valuable commodity that they have is publications, and so they publish. The moment keeping it secret is more valuable to the business than publishing, they will stop publishing, or they'll publish other stuff, but not the interesting stuff. So, the cynic in me says they're very free and open in publishing because they ain't got the goods. Obviously I'm in favor of everybody publishing. I'm a physicist, I like publishing good results and reading them. But yeah, I think you can tell the things are still immature because they're still publishing what looks like their best cutting edge results. And of course you never know what they don't publish.

Yuval: So that's kind of a litmus test. When they stop publishing, you know now it's money time for quantum.

Tom: Or when they start publishing and you read it carefully and you said, yeah, who cares? Then you kind of know that they're holding back.

Yuval: Looking at your crystal ball, what do you think is going to happen in 2022 in quantum? What are you hoping that will happen? What do you think is realistic that will happen?

Tom: Honestly, I think the status quo on the straight line of what we've been seeing. What we've been seeing is very impressive. I don't think that 2022 holds anything earth shaking. Learning how to do interconnects and scaling in transmon stuff. Again, folks like PsiQuantum who are going for the million qubit chip in a tier one foundry, they're on schedule, they're not going to have anything next year. So I think people will slowly come to understand that algorithms are going to be hardware specific. And when they get their heads around error, correction, and fault-tolerant channels, algorithms are going to be fault tolerant channel aware because you're going to want that factor of two or four faster throughput, because you want to cross it. There's a threshold you want to cross, and if you can come 90% of the way to what a Linux cluster can do, who cares? If you can do 200% or 400% of what a Linux cluster can do, the world will pay attention.

I know folks, a friend of mine works in LIGO and Columbia, programs in C++, but still writes assembler, because you got to. So that's what I see. I've been saying that this hardware agnostic stuff is all very wonderful, but that is way out in the future. And hardware aware programming and compiler assisted, hardware aware, stuff is what people are going start to focus on or should at least. Whether they will, I don't know.

Yuval: Putting the physicist aside for a second, how much do you think people are paying attention to the hardware? Because you could get caught up in the news cycle. Oh, I've got more qubits. I've got this better way. I've got this modality, that modality. Do you think enterprises are ready to commit to one vendor or one particular type of modality, or they're still holding back and say we need the hardware agnostic because we don't know who's going to be the winner.

Tom: Oh, forgive me, I hear that as two separate questions. Again, the wishful thinkers would like to simplify their job and make it somebody else's problem. And a lot of people will go forward that way, but again, the one who figures out how to cross the line and offer something to their customers, that's when people will realize, and I think that that's what's going to happen. So I just don't see hardware agnostic being the winner. And I think that a lot of people will cling to that model because, hey, it makes their life easier. People are wrong in business all the time.

Yuval: If the hardware plans hold, and we'll see in the next few months, or certainly next year, computers with 100 qubits and better noise characteristics and so on, to the point where they can no longer be simulated on classical machines. Does that excite you? Does that scare you? How are you thinking about that?

Tom: It definitely excites me, but again, there are strong limits to what you can do in NISQ. 100 noisy qubits is better than 50 noisy qubits, but then, exactly what I was saying, figuring out what I can actually do that's useful. Again, this is pure physics and not the commercial market, but a group at University College London figured out that you can indeed do good physics computations with the noisy, random circuits of the kind that Google presented. And that has a computational power that you can extract, but that relies on something called quantum typicality and physics systems tend to be able to take advantage of quantum typicality. As far as I know, there's no algorithms in financial or logistics or anything where you could apply that kind of concept. So here you have an application, and that one might grow. I mean, the physics modelers might take off in 2022, and I'll be happy for them, but you know, so I'll read papers in Phys. Rev., Hey wonderful. That excites me as a physicist, but that's not why I am trying to be a quantum champion in our sphere.

Yuval: You've been around for a while and you've seen new technologies come and go and sometimes take off and sometimes not take off. How much are you worried about the possibility of a quantum winter where it's just over-hyped, doesn't deliver, and people put it in the drawer for 10 years.

Tom: I'm not personally concerned about that. It is clearly overhyped, but as the joke goes, just because you're paranoid doesn't mean they're not out to get you. Just because everybody is over-hyping it doesn't mean that there aren't nuggets there, and that's exactly what I was saying earlier. They will be niche applications. They will not replace your entire data center, but they might power one or two products in the first few years. And then you have something that nucleates the outward growth. So, no, I am not concerned about the quantum winter at all. I was famously wrong about gallium nitride. We were working on two six blue lasers, and I looked at a piece of gallium nitride and said, "This material is crap. You're never going to make an optical device out of this." And of course Nakamura got a Nobel Prize for proving me wrong. So no, I'm not concerned.

Yuval: So hypothetically, you're the master of the universe now, intergalactic dragon slayer, what would you have us hardware and software vendors work on for the next 18 months?

Tom: Again, I happen to think that, well, PsiQuantum is going forward, and I think that that's a great technology. I think the ultracold atoms is an area that is not getting enough coverage because, even without error correction, they're pretty quiet. And again, an array of 100 by 100 individually addressable atoms in laser traps is old technology. I mean, I've seen that with my eyes in a lab and you can casually dump the entire array and 30 seconds later it's back. So I think that the ultracold atom folks are going to rise in prominence, and I honestly don't know if the transmon folks are going to get their interconnect problems solved, because, again, my own ignorance is not a guide for anything, but I am really glad that I'm not the person who has to figure out how to get a million microwave pulses into a cryostat and not have the crosstalk totally kill you. Again, I'm not the world's expert at that. I am sure people are making great progress, but that seems like a very steep hill to climb.

And... long-term: photons, it's hard to argue with Silicon fab technology and photons. If somebody could show actual MBE in silicon quantum dots, all DC voltage addressed, that'd be great. That would even be optical, but you got to look for the technologies where the scalability is there and the innate noise characteristics are either livable or completely solvable.

Although the fact that I like the math, no connection to PsiQuantum, I'm not advertising for them, but everything that they're doing seems to be the right thing. They're saying, here's how you build a fault-tolerant channel, oh, it happens to be topologically exactly isomorphic to the surface codes that everybody else is talking about, and you're going to do it this way, so you're focusing on the right problem. Excellent. I'm wondering, the answer to your question is people are going to focus on that which delivers actual value. They're going to start cutting through their own hype and realizing that you have to bring something to market to sell it, and you're going to need those technologies that will actually behave well and scale.

Yuval: Excellent. Tom, that was very interesting. How can people get in touch with you to learn more about your work?

Tom: The email is tmarshall2@bloomberg.net, or the somewhat easier thomas_marshall@msn.com, one of the oldest domains that there is.

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

Tom: Thank you, it's been my pleasure, and I look forward to talking with you more.




My guest today is Tom Marshall, software engineer from Bloomberg. Tom and I spoke about deploying quantum computing in data centers, his assessment of various quantum modalities, killer applications for financial services and much more.

Listen to additional podcasts here

THE FULL TRANSCRIPT IS BELOW

Yuval: Hello Tom, and thanks for joining me today.

Tom: Happy to be here.

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

Tom: My name is Tom Marshall, I'm an experimental physicist. I'm doing software architecture and security for Bloomberg, and I've been paying a lot of attention to quantum computing.  And all of my opinions are my own.  I'm not speaking for Bloomberg here in any way.

Yuval: So, are you more on the software side, or more on the physics side, or some combination of both?

Tom: Well, my day job is purely software, but, and again, I'm a physicist by education and nature, so you can't take that out of me.

Yuval: As am I. I have a master's in physics, but I don't do physics on a daily basis. Let me ask you a software question. We see companies doing proof of concepts, and some of them are more successful than others. And some people are saying, "Okay, what would it take to move these proof of concepts from the sandbox to a production environment?" How do you see quantum fit in enterprise computing architecture? I mean, would the IT folks manage it just like they manage cloud resources? What do you see as the issues of integrating quantum into enterprise computing architecture?

Tom: It's a good question. I think it's a lot more complicated than that. First of all, I always like to say, there is no cloud; it's just somebody else's computer. So, where the physical hardware is located is not too terribly relevant to typical commercial end-users. But the issue, as I see it, is from the moment something can be done that is of commercial interest to someone, until the time that it is actually a commodity and we can say, how do we want to deploy it? And should AWS host it, or should we buy our own? Or whatever. I think that there will be a period of scarce resources and quite a bit of disruption in industry, and different industries will react differently. And again, the financial services industry, I think, will be especially volatile. So I think it's going to be a mess, and the physical deployment is the least of anybody's concerns until it becomes a commodity, and then the data center people will figure out how to manage it.

Yuval: If the physical location is not the issue, what is the issue? Is it integrating it into data input? Is it putting it in a workflow? Is it a service level management issue? What is the issue?

Tom: Scale. I think that, for instance, if PsiQuantum is successful, they have a technology that will scale relatively quickly. I am not as comfortable that the transmon superconductor machines will scale as friendly as you like. You can build a lot of them and they'll work, but they won't be 24/7 commodity reliable, I think. They will be expensive to maintain, and therefore, I think they'll be a scarce resource for times frame of years until they get it worked out because it'll do something useful, it will be very valuable, but you won't have a million of them in a data center. That's what I see. I'd like to be wrong on that because I'm not in favor of disruption, but I see disruption in the future.

Yuval: So if you want to run portfolio optimization, for instance, just to say, common financial services use case, how often do you run it? How, how many times a day, how many times a second do you run it?

Tom: I do know that it's compute limited. I do know that, could it be done in real time, that's a service that could be offered to day traders going to market with a balanced risk portfolio thing in and out in real time. I mean, the poster child is derivatives pricing, very slow, very expensive. If you could just price the basket of derivatives and in and out, that would, I think, revolutionize the market. Again, I'm not a financial guy, but it seems that we're so compute limited that that's where the disruption will be.

Yuval: I agree with you. The point I was trying to make, or trying to ask is: some computations need to happen in real time or constantly through the day, and others could happen in batches. Maybe it's not a financial services question. Maybe it's a shipping company that needs to recalculate the routes once a day or a couple of times a day. And so perhaps the 24/7 availability of compute resources is not as critical for some applications.

Tom: Oh, I completely agree, and those are somewhat less disruptive. I don't know how Amazon handles logistics, but let's just say that Amazon could greatly improve their logistics. Okay, fine. So my socks are two pennies cheaper, or Bezos gets two pennies richer, but the world doesn't change because you're doing batch computations there. So I hear what you're saying. I simply haven't focused on those applications. They're probably good, I mean, they're important, but that's not where I see the biggest social impact and economic impact on quantum computing.

And again, something that we've talked about, if you could say the so-called Greeks, the partial derivatives of risk and value, if you could calculate them in real time would be great, but if you cached them once a day and then have classical algorithms hit that cache, that's great, that's wonderful. Again, it will be opportunistic, it will not be that some product manager specifies what they want, it'll be, well I can give you this. Oh, good, I'll run that once a day, and then get the cache. Or, oh I can do this in real time because I can. So anyway, that's the sense of how I'm reacting to your comments.

Yuval: I understand, but it seems that there are many potential use cases in financial services, some more disruptive than others. And I think that there's a consensus amongst financial services companies that quantum holds opportunities or potential opportunities for true breakthroughs. In that context, we see many financial services firm publish results of research in academic papers. Is that just a recruiting strategy? Do you expect that to continue into the future, if indeed you have some true competitive advantages in quantum?

Tom: Well again, this is my cynical view of the world, but right now the most valuable commodity that they have is publications, and so they publish. The moment keeping it secret is more valuable to the business than publishing, they will stop publishing, or they'll publish other stuff, but not the interesting stuff. So, the cynic in me says they're very free and open in publishing because they ain't got the goods. Obviously I'm in favor of everybody publishing. I'm a physicist, I like publishing good results and reading them. But yeah, I think you can tell the things are still immature because they're still publishing what looks like their best cutting edge results. And of course you never know what they don't publish.

Yuval: So that's kind of a litmus test. When they stop publishing, you know now it's money time for quantum.

Tom: Or when they start publishing and you read it carefully and you said, yeah, who cares? Then you kind of know that they're holding back.

Yuval: Looking at your crystal ball, what do you think is going to happen in 2022 in quantum? What are you hoping that will happen? What do you think is realistic that will happen?

Tom: Honestly, I think the status quo on the straight line of what we've been seeing. What we've been seeing is very impressive. I don't think that 2022 holds anything earth shaking. Learning how to do interconnects and scaling in transmon stuff. Again, folks like PsiQuantum who are going for the million qubit chip in a tier one foundry, they're on schedule, they're not going to have anything next year. So I think people will slowly come to understand that algorithms are going to be hardware specific. And when they get their heads around error, correction, and fault-tolerant channels, algorithms are going to be fault tolerant channel aware because you're going to want that factor of two or four faster throughput, because you want to cross it. There's a threshold you want to cross, and if you can come 90% of the way to what a Linux cluster can do, who cares? If you can do 200% or 400% of what a Linux cluster can do, the world will pay attention.

I know folks, a friend of mine works in LIGO and Columbia, programs in C++, but still writes assembler, because you got to. So that's what I see. I've been saying that this hardware agnostic stuff is all very wonderful, but that is way out in the future. And hardware aware programming and compiler assisted, hardware aware, stuff is what people are going start to focus on or should at least. Whether they will, I don't know.

Yuval: Putting the physicist aside for a second, how much do you think people are paying attention to the hardware? Because you could get caught up in the news cycle. Oh, I've got more qubits. I've got this better way. I've got this modality, that modality. Do you think enterprises are ready to commit to one vendor or one particular type of modality, or they're still holding back and say we need the hardware agnostic because we don't know who's going to be the winner.

Tom: Oh, forgive me, I hear that as two separate questions. Again, the wishful thinkers would like to simplify their job and make it somebody else's problem. And a lot of people will go forward that way, but again, the one who figures out how to cross the line and offer something to their customers, that's when people will realize, and I think that that's what's going to happen. So I just don't see hardware agnostic being the winner. And I think that a lot of people will cling to that model because, hey, it makes their life easier. People are wrong in business all the time.

Yuval: If the hardware plans hold, and we'll see in the next few months, or certainly next year, computers with 100 qubits and better noise characteristics and so on, to the point where they can no longer be simulated on classical machines. Does that excite you? Does that scare you? How are you thinking about that?

Tom: It definitely excites me, but again, there are strong limits to what you can do in NISQ. 100 noisy qubits is better than 50 noisy qubits, but then, exactly what I was saying, figuring out what I can actually do that's useful. Again, this is pure physics and not the commercial market, but a group at University College London figured out that you can indeed do good physics computations with the noisy, random circuits of the kind that Google presented. And that has a computational power that you can extract, but that relies on something called quantum typicality and physics systems tend to be able to take advantage of quantum typicality. As far as I know, there's no algorithms in financial or logistics or anything where you could apply that kind of concept. So here you have an application, and that one might grow. I mean, the physics modelers might take off in 2022, and I'll be happy for them, but you know, so I'll read papers in Phys. Rev., Hey wonderful. That excites me as a physicist, but that's not why I am trying to be a quantum champion in our sphere.

Yuval: You've been around for a while and you've seen new technologies come and go and sometimes take off and sometimes not take off. How much are you worried about the possibility of a quantum winter where it's just over-hyped, doesn't deliver, and people put it in the drawer for 10 years.

Tom: I'm not personally concerned about that. It is clearly overhyped, but as the joke goes, just because you're paranoid doesn't mean they're not out to get you. Just because everybody is over-hyping it doesn't mean that there aren't nuggets there, and that's exactly what I was saying earlier. They will be niche applications. They will not replace your entire data center, but they might power one or two products in the first few years. And then you have something that nucleates the outward growth. So, no, I am not concerned about the quantum winter at all. I was famously wrong about gallium nitride. We were working on two six blue lasers, and I looked at a piece of gallium nitride and said, "This material is crap. You're never going to make an optical device out of this." And of course Nakamura got a Nobel Prize for proving me wrong. So no, I'm not concerned.

Yuval: So hypothetically, you're the master of the universe now, intergalactic dragon slayer, what would you have us hardware and software vendors work on for the next 18 months?

Tom: Again, I happen to think that, well, PsiQuantum is going forward, and I think that that's a great technology. I think the ultracold atoms is an area that is not getting enough coverage because, even without error correction, they're pretty quiet. And again, an array of 100 by 100 individually addressable atoms in laser traps is old technology. I mean, I've seen that with my eyes in a lab and you can casually dump the entire array and 30 seconds later it's back. So I think that the ultracold atom folks are going to rise in prominence, and I honestly don't know if the transmon folks are going to get their interconnect problems solved, because, again, my own ignorance is not a guide for anything, but I am really glad that I'm not the person who has to figure out how to get a million microwave pulses into a cryostat and not have the crosstalk totally kill you. Again, I'm not the world's expert at that. I am sure people are making great progress, but that seems like a very steep hill to climb.

And... long-term: photons, it's hard to argue with Silicon fab technology and photons. If somebody could show actual MBE in silicon quantum dots, all DC voltage addressed, that'd be great. That would even be optical, but you got to look for the technologies where the scalability is there and the innate noise characteristics are either livable or completely solvable.

Although the fact that I like the math, no connection to PsiQuantum, I'm not advertising for them, but everything that they're doing seems to be the right thing. They're saying, here's how you build a fault-tolerant channel, oh, it happens to be topologically exactly isomorphic to the surface codes that everybody else is talking about, and you're going to do it this way, so you're focusing on the right problem. Excellent. I'm wondering, the answer to your question is people are going to focus on that which delivers actual value. They're going to start cutting through their own hype and realizing that you have to bring something to market to sell it, and you're going to need those technologies that will actually behave well and scale.

Yuval: Excellent. Tom, that was very interesting. How can people get in touch with you to learn more about your work?

Tom: The email is tmarshall2@bloomberg.net, or the somewhat easier thomas_marshall@msn.com, one of the oldest domains that there is.

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

Tom: Thank you, it's been my pleasure, and I look forward to talking with you more.




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.

See Also

No items found.

量子ソフトウェア開発を開始

お問い合わせ