Loraine Lawson
Good day. Welcome to the Buzz, a Bank Automation News podcast. I’m Editor Loraine Lawson and recently I spoke with Sam Mugel, the CTO of Multiverse Computing, which started as a fintech specializing in quantum computing. Early customers include BBVA and the Bank of Canada. Mugel also holds a Ph.D in quantum computing and was instrumental in developing Multiverse’s Singularity,a spreadsheet app for quantum investment optimization that makes makes quantum algorithms accessible. I asked him to explain what makes quantum computing unique.

Sam Mugel
So in standard, what we call classical computing, that’s really well adapted to doing certain operations, like for instance, additions and multiplications are very, very easy for classical computing. And then Quantum Computing is well adapted to do a a whole different set of operations. So you can think of quantum computing as we’re using different basic laws to make calculations. And in the future, we’ll probably see that classical and quantum computing actually very compatible, basically. So we’ll use classical for some things and quantum for some things. The type of things that quantum computing is well adapted to doing is, for instance, looking for an item in the list. If you take the phonebook and you want to find a specific phone number in it, this is actually really difficult to do with normal computing, you have to look at every single phone number until you find the right one. And one of the really big victories of quantum computing, one of the first victories that really started off the field, was we realized that so we have a very good understanding of how much time it’s going to take us with classical computing to find that phone number. And we realized that with quantum computing, we could beat that time by quite a lot by root to the number of the number of phone numbers you’ve cut. So if you’ve got a very long list, that’s going to be really significant, and might be, for instance, part of the reason why Google is so interested in investing so much in quantum computing. Quantum Computing is also good at things like training machine learning models, which right now is costing an awful lot of energy and awful lots of our computational power so it’s really good at things like really difficult optimization problems, which are everywhere, particularly in finance. So that’s kind of a

Loraine Lawson
I have a sort of silly question, I guess. But like, how big is a quantum computer? Is it sort of like back in the old days when you know, mess? My cell phone now is the power of what used to be a roomful of computers? Are we at that stage with common community? Computers? Are they relatively small?

Sam Mugel
Yeah, so we’re kind of like, physicists like to think that we’re in the 40s, or 50s, of classical computing with quantum computers. So your regular quantum computer is about as big as this room. So for those of you that can’t see this woman, it’s about three meters by three meters. And most of that space is usually a huge fridge. Most quantum computers need to be kept very close to absolute zero. So that’s minus 270 degrees Celsius. And your chip where the computations are happening is actually quite small. And then all the rest of it is this huge fridge.

Loraine Lawson 08:00
Wow. So what makes it a quantum computer versus a regular computer? Is that too big of a question?

Sam Mugel
No, really not. So we’re in the middle of what we call the second quantum revolution. The first quantum revolution was everything like lasers and semiconductors. These things use quantum physics. Quantum physics is the science that controls electrons and atoms and light particles. So you need that science to be able to make laser to be able to make a semiconductor. There is an additional element, though we’re now using in what we call the second quantum revolution. And this is let’s now have quantum properties over many elements in the system. So we’re not using quantum properties to produce one photon, or using quantum properties to control how one electron flows through a semiconductor, we’re now going to use, okay, this bit of information over here is bound somehow to another bits of information. And that’s going to allow us to do calculations faster. So this property is called correlation. And this is really the where we’re really using these this new resource, basically, to do calculations in quantum computers. So it’s

Loraine Lawson
very different than what we have now on our desktop. It sounds like in terms of how it works.

Sam Mugel
Maybe it’s useful to think of it as actually not that different in a, in a classical computer, you have a bits of information that can be zero or one. And then when you apply an operation, you can apply binary gates. So things like ands, and ORs nots. And by composing these things, you’re able to do calculations. And in quantum computing, you also have bits of information. But we use quantum properties so that these can be simultaneously zero and one at the same time. This is a weird thing, that we don’t know why it happens. We, we’ve just observed that in quantum physics, this type of thing can happen. And then I still play operations on my, on my bits, like knots and O’s. And when I apply that operation, I’m doing it simultaneously to both pieces of information, both the zero and the one. So I’ve created essentially a parallel computing machine that implements that thing naturally. The big difference is, for one part, I’m able to do these operations in parallel natively, which isn’t possible on a normal computer. And for another part, the gates that we have are actually different. So I don’t have an end gate on my quantum computer. Instead, I’d call it something else. That’s called a topic.

Loraine Lawson
Let’s see. And so does that mean that that you use the key use the same programming languages on a quantum computer, then?

Sam Mugel
Yes, and no? That’s a good question. Okay. So we interact with when when we write a code for a quantum computer, we write it in Python. So Python, very standard, programming language that’s available to everyone. So the way we interact with the quantum computer is the same as in we’re using the same language. However, what that language is actually doing is on the level of the quantum computer, it’s controlling, which laser is going to shoot at which time and which bits do you want to, like, carry around to where and things like this. And so we’ve used Python to create new programming language, new programming languages, examples of this are qiskit, Cirque pennilyn, and these programming languages act as an interface between Python and Python, and the actual computer.

Loraine Lawson
I see. So I know banks are pursuing quantum computer or some of the bigger banks, do you have any idea what they’re pursuing it to what in? Or maybe it’s a better question just to say, what are the applications for financial institutions, particularly when it comes to automation with quantum computing?

Sam Mugel
Absolutely. Yeah, this is a great question. I think many banks have the same problems at the end of the day they struggle with, okay, if I issue a loan to a customer, what’s the risk associated with that loan? What’s the risk of that? My customers credit score is going to decay massively over the lifetime of the loan. And these are things that you tackle through machine learning. And there’s lots of examples like this. So problems associated to a credit score. Going through detection to anti money laundering to customer propensity, if I offer my customer, Bob a credit card, what’s what’s the likelihood he’s going to, to accept to start a credit card plan? So these are very, very difficult, very high worth problems for all banks struggle with. And the underlying machine learning algorithms, we know we have lots of evidence that you are able to train these faster on quantum computers. And so there’s a lot of interest in that direction, essentially. So this ties in to your automation question. The other areas the banks can confine loss of value from quantum computing is optimization in finance, things like, again, portfolio optimization, or index tracking, collateral, …, optimization, portfolio stress testing, there’s a whole like wealth of problems out there that are extremely high value that where banks can extract a lot of value from quantum computing. Finally, and maybe the most, like ambitious and exciting area is everything that has to do with pricing applications and market forecasts. The way we usually do this is we’ll make a lot of simulations, random simulations of what the market is going to do. And then make like, use averages overall of those simulations to make predictions. This is extremely intensive, from a computing point of view. Banks very often run computations that more than 24 hour long term to price exotic options. And, again, in this domain, there’s lots of evidence that this might be something that quantum computers can can do much, much faster, essentially.

Loraine Lawson
And you work, you’re the CTO of multiverse computing. Can you explain a bit about multiverse computing and your work with quantum computing there. And then I know you have big clients, how are they using your product?

Sam Mugel
So Multiverse computing, we started out as a as a FinTech, essentially. So we said, let’s tackle really difficult problems in finance, and let’s do it using quantum computing. And using a classical algorithm, which we call quantum inside, called Tensor Networks, which is different from TensorFlow. And now, so we started out three years ago, and now we’re where stage where we’re branching out a little bit, we’ve seen lots of areas where we can add value in quantum computing. And we’ve had lots of big clients. We’ve just announced yesterday, actually, some work that we’ve been doing with Bank of Canada from past 10 months. That was an extremely interesting project where we were looking at how well quantum computing affects our ability to predict the effects of regulations for for cryptocurrencies. We’ve worked with a lot of banks, we’ve worked with BBVA acacia, which are two major Spanish banks. We’re working with Claudio he called. So that’s the world’s 11th biggest bank, we’re also working with American banks. And then recently, and and the Spanish tax agency that was really interested in project and tax fraud detection. And recently, something really interesting happened where players from other fields that’s approaching us and said, hey, you’ve got this great portfolio optimization solution, we actually think that this can be applied to optimizing energy markets. Turns out, both problems are very similar. So who do I buy my energy from? How much do I sell it for? And things like this. And so that’s a very successful project that we’ve been doing with a major petro chemical and energy management company called Repsol in Spain. And we’ve recently also started a multi multi million euro project with Bosch in smart manufacturing. So there’s some extremely interesting problems we can tackle there, especially that have to do with predictive maintenance.

Loraine Lawson
So what makes a problem good for quantum computing over regular computing, what are the hallmarks of what you look for in a project, so

Sam Mugel
you can have one four things simultaneously. First, you need your problem to be quite a small input, the reality of quantum computers at the moment is the the processes are still quite small we, we have for the largest processes 1000 bits to play around with. So your inputs have to be quite small, you want there to be a lot of intermediate states. So for the portfolio optimization example, for instance, there’s a very large number of portfolios that you can that you can build, right. And so this is this is going to, like the so the portfolio optimization state would fulfill that condition. You also want it to be a very high value problem. And ideally, something that you’d have to run very often. The thing is, though, quantum computing is very expensive. And so you need to justify that investment. And finally, you need it to be a problem that’s fair, hard to solve classically. So in the ideal case, for instance, Isa, or looking for a number in the list, your best solution is to brute force it. I’m just going to look at all options until I find the right one. And so this is a really well suited problem for to tackle with quantum computing.

Loraine Lawson
So as part of multiverse computing, do you actually have Have the quantum computer yourselves and they sort of uses as a service or how does that work.

Sam Mugel
So we don’t own any quantum computers, we’re, we’re a software company, we build software for our clients and our software relies on many different backends. So we have partnerships with all the big quantum software companies out there, from like D wave to IBM, we have partnerships with Microsoft AWS with Pascal is a really good partner of us. Inq, Alexander. And we have access to all these quantum computers to the cloud. And a big part of the value that will bring is when the client approaches us with a specific problem, we’ll tell them out, your problem is actually amenable to be solved with this particular solution. And the best hardware to run it on is this particular piece of hardware, and we’ll develop a solution. And when I say like this, it sounds like we’re a service company. But there’s a whole whole part afterwards where we take that solution and integrate it into our toolbox. So we’re, we’re a SaaS company, and, and we’re working on developing this big tool that we can then license out to many customers.

Loraine Lawson
Okay, and, and the banks have they had you had? Has there been commonality between what they’ve had you work on? Are they all different? Can you talk about any of those projects?

Sam Mugel
Absolutely. So with BBVA, and Keisha, we’ve been doing lots of work on portfolio optimization. This case, this is an interesting problem, especially when you’re going to start adding, okay, maybe I don’t want to just buy some assets and hold them, maybe I’m actually interested in changing my holdings would take. And then it becomes a very, very difficult problem. The example of that is for instance, okay, I know that. So if I’m just saying name, like Fidelity of Blackrock, I know that my customers want to buy this particular I need to constitute this particular portfolio. However, if I buy everything in one go, this is going to cause a significant market impact. And I’m actually going to lose lots of money from that. So I’ve got my target portfolio. And then I need to I need to through a sequence of buys and sells get to that target portfolio. And the the optimization problem to get to that target portfolio, while reducing market impact is actually a very, very difficult one. Aside from that, with Kinsey Agricole, so once again, was loving the biggest bank, we’ve had two projects on the go with them, one of them has been, has had to do with this credit scoring problem. So let’s identify which of the customers are actually going to stop being investment grade over the last lifetime, we’ve alone. The other one is an exotic option pricing problem that we’re tackling with Tensor Networks, these two are very much work in progress. We’re really getting some some very exciting results. So I’d expect that in the next two or three months, we should see some fairly high impact publication coming out of this.

Loraine Lawson
So if you could say one thing, or, you know, one general statement to banks, about this technology, what would you want them to know? It doesn’t have to be just one.

Sam Mugel
No, I like the question. I think I think quantum computing, one thing that I’ve seen from talking to a lot of bank says that generally, banks are very excited about accelerating Monte Carlo calculations. And I think that there’s a missed opportunity here because as a bank, what you want to do is not to accelerate whatever like calculation you’re using. So MonteCarlo is a is a very used and very general method for solving lots of problems. And you’re not actually interested in doing that you’re you’re actually interested in solving a particular problem and, and you know that you use Monte Carlo to solve that particular problem. Like for instance, you want to create a Bermuda option. and you’re going to use Monte Carlo for that when you see Monte Carlo is taking lots of your computational capacity. And in lots of cases, you’re, you’re better off trying to think about what other means, if you’re thinking about applying quantum computing to this problem, rather than being stuck on the idea of accelerating Montecarlo, you’re better off thinking, what other ways are there, pricing, removed options? And then, and then seeing where quantum computing can have an effect.

Loraine Lawson
You’ve been listening to the Buzz, a Bank Automation News podcast. Thank you for your time, and be sure to visit us at Bank automation news.com for more automation news. You can also follow us on Twitter and LinkedIn. Please don’t hesitate to rate this podcast on your podcast platform of choice.