Understanding quantum computing’s potential in financial services with Goldman Sachs’ Will Zeng and IBM’s Stefan Woerner

What is quantum computing?

Will Zeng, Head of Quantum Research, Goldman Sachs: Broadly, it's a new way of making much more powerful computing hardware. So all the hardware that we interact with today, your Intel, your iPhone, everything, it's based on physics. But it's based on the rules of electromagnetism, which we've known about since the 1800s. And in the 1900s, the big revolution was understanding that underneath the electromagnetic operating system, there are quantum mechanical rules. And the idea behind quantum computers is to take advantage of these features and this new power that's in quantum physics, and make computing technology based on those new rules. And it turns out that, in a lot of cases, it opens up radically more powerful kinds of computing.

IBM and Quantum

Stefan Woerner, Quantum Applications Research & Software Lead, IBM: IBM covers the full stack for quantum computing. We have a large effort building the hardware, and building these quantum computers is very difficult. And this is really like taming nature at its extreme. We are building this hardware up from the foundational research to now the first systems that we offer for our commercial clients over the cloud.

These systems are still at a size where they are used for research. Since they are so different, you need to learn how to use these systems before you can really leverage them. So learning how to use them makes sense right now before the large scale systems are available that run a productive workload. So that's one side: we did the hardware research and development to make this available to our partners.

Then on top of that, we also develop the control software of the application software, like for optimization and things like that. We have software tools that people can use, whether that's quantum researchers that look into how to improve algorithms, or whether that's application and subject matter experts that want to see what an interface looks like, how things work at a small scale. We really want to understand the applications for different industries. What are problems that we are struggling with today, solving them with classical hardware, where quantum can help? This is a project in an area we collaborated with Goldman Sachs and at the end of last year, we published this first joint paper.

SPONSORED

IBM's quantum computing roots

Stefan Woerner: IBM has been working on quantum computing for a long time -- I think starting in the 80s or 90s. So really like for decades, laying the theoretical foundation. Some of our colleagues, like Charlie Bennett, they really were the pioneers of quantum computing. And since then, we contributed to that through all these years on different levels. Of course, in the early years, this was really on the foundational level from the theoretical point of view, but also foundational hardware research. And since 2016, we have made the first quantum computer available through the cloud for the public and for our partners.

Quantum computing and modern finance

Will Zeng: One of the reasons that finance is so exciting for applying new kinds of commercial computing technologies is that we have a lot of mathematical problems that are pretty easy to specify and that there's been a lot of incentive to work with classical computing hardware, non quantum. The old generations of hardware have been optimized aggressively in a lot of cases. So there's a real need for new thinking.

There are three broad categories that we're looking at right now. The first is in what I call broadly simulations. The paper that Stefan alluded to that we worked on was for the pricing of derivatives. The math setup is calculating expectation values of stochastic processes, functions, processes, which comes up all the time in finance around this modeling.

The second is optimization. There are a lot of hard optimization problems across financial services, portfolio optimization being maybe the most obvious one, but they come up in all sorts of places.

The third category is machine learning. And here, there are applications in trading, but there's also applications in things like anti money laundering, or avoiding fraud. And those three categories are already really, really large. And so what our research group is doing is trying to pick out more specific benchmarks where we think that quantum computing could be most useful first.

Quantum's potential for improvement

Will Zeng: We still have just prototypes. Some of the optimization use cases that we're excited about are ones where we have strong theoretical proof of big possible improvements. So, for example, in derivative pricing, at a theoretical level, you can show very generically that you can get tens, hundreds or thousands of decks sped up in certain kinds of theoretical models for how to do the problem. Then what we do is try to take that theory and apply it to machine learning.

Classical machine learning hasn't totally penetrated finance yet. Regular ML is still spreading throughout financial services. And so that means that the benchmark is a little more vague. With the first few generations of quantum computers, we have less of a theoretical handle on how much better they might be. It's a really, really active area of research. But it's a bit more nascent.

When will we see more than prototype quantum computers?

Stefan Woerner: At the moment, we have the first devices that have like 65 qubits. For example, this is the largest device that we make available over the cloud today. And over the next couple of years, we plan to reach the first devices with like 1000 qubits or more. But here we talk about noisy qubits.

Some of the algorithms that Will mentioned, where we have this strong theory behind, they require what we call a logical qubit. This can be in the in the hundreds or 1000s of physical qubits per logical qubit. So here, we certainly have a little bit of work to do. We hope that around 2023, when we reach the first chip with 1000 plus cubits, we can start studying this, demonstrating this first logical qubit. Then we can scale within this decade to large scale quantum computers that can demonstrate quantum advantage within the state this decade.

If you look at the hardware roadmap that we published, this is the way forward and we hope that by the end of this decade, we really have these large scale quantum computers that can run error corrected workloads. There's already quite a wide variety of suggestions on algorithms for our current noisy devices, whether this is on machine learning or optimization, and we are investigating what can be done with those and where can we expect some speed ups or some performance improvements. It will also be very interesting in the coming years to see what we can do with these smaller scale quantum computers in these different areas.

Investment in quantum computing

Will Zeng: It might be worth mentioning, when you think about where the field is going, the first prototype quantum computers have only been available in the last couple years. When Stefan and I started, it was just a thing that some academics and some deep research groups like IBM did. Now, there's been more than $20 billion of global government public research funding announced over the last few years. Venture capital has invested almost $1.5 billion into the space. The growth and trajectory is really encouraging. We see a lot of innovation happening.

Derivative pricing

Will Zeng: Derivative pricing is obviously a really, really ubiquitous use case. And it's a very concrete use case, where we know that there's some theoretical speed up from quantum computers. It's what's called a quadratic speed up. So that means that if I want to make my simulation on a normal computer more accurate by a factor of, let's say, 100, that I need to use 100 times more classical compute power, but I would only need to use 10 times more quantum computing -- that kind of scale.

That's still in a theoretical model. What we wanted to do with this research, in collaboration with Stefan, was to take those theory numbers and make them concrete. We wanted to say, as far as we understand it today, what are the real specs of the quantum computer so that we wouldn't need four versions of derivative pricing that my colleagues in global markets would care about? And so those specs are things like the amount of quantum memory, the clock rate of a quantum processor, and the error rate. Stephen alluded to the error rates being kind of high. And so what is the actual target error rate that you need? And in doing that, we learned a lot.

We learned that a lot of theoretical work in the literature actually has some gaps and holes. So, we had to come up with some new algorithmic tricks that we can talk about. But it also led us to some first concrete estimates that we see as a roadmap that we can now go talk to the hardware vendors, look at their roadmaps, see how they line up, and then also prioritize our theoretical work to compile down, really to reduce the time to when the real potential of this application can be realized.

Practical takeaways

Will Zeng: Today's hardware specs are not good enough yet. There are a few milestones to go through. For example, for the benchmarks we picked, it looks like you need 7000 to 9000 logical qubits. And today's hardware is around 65 -- it will soon be at 100. It's growing rapidly. But it's not quite there yet. So from a business side, the takeaway is we're not there yet.

The second takeaway at the business level is that we actually could do a wall clock time, concrete estimate of what the specs are. And I think that's causing a shift in the discussion around applications of computing, which I think, until now, has been, oh, there's some theory. We're hoping that as prototypes get developed, we can figure out how to use them. Here, we've been able to be really concrete and specific, for an end to end calculation from top to bottom, and estimate how long it would take. Stuff like that speaks to the maturity of the field.

Democratizing access to quantum resource

Stefan Woerner: I think that's what we are already doing, right? We are building an ecosystem around quantum computing. We're making this accessible through our IBM quantum network, where we already have 100 plus partners and members, and all of them can access this. We even have quantum computers available through the cloud for free to the public. But those are not the premium devices. So I think building the ecosystem and having access through the cloud to a large group of users is very important for multiple reasons. Particularly in the process of getting to useful applications, we really want and need a variety of people that look into what can be done with this, that look into it from completely different angles and industries, and help to grow this.

IBM has very much been pushing the open source development of the underlying software: our software framework that allows to write quantum circuits that can be run on the hardware, but also to provide the algorithms on top. Even more applications, like optimization or machine learning, this is all open source, and this is a community effort. And the community contributes to that. And together we build this. So I think having broad access is very important to grow this whole industry and to learn what can be done to identify the applications that really profit from it.

Making quantum a reality

Will Zeng: That's my job. When I listen to stuff about the ecosystem, it's like where machine learning was 10 or 12 years ago. Like where the compute power is not quite there yet. But as quantum hardware matures, and there's a lot of roadmaps out there and a lot of money going into starting to build these machines, big ecosystems will pop up. New applications will continue to be found. We are, in some ways, here to champion that financial services is a really great place to start applying this tech. It's got concrete problems, got global access to deploy new kinds of compute solutions. And so, part of this work is to sort of trumpet that it's a good place for quantum computing to develop.

There's maybe 150 companies in the ecosystem if you think of quantum tech a little broadly. There's a whole stack of things here, like people who make hardware, there's people who make control systems for the hardware, people who make software or applications research. And that's really global. You see a lot of the big startups in the US but also in Europe, Australia, Japan, Netherlands, France, Germany, and the UK. It's starting to emerge into real industry.