Transcript | From Astrophysics to Quantum Computing Algorithms: A Software Engineer’s Journey

One of the hottest jobs in quantum computing in the coming years will be that of software engineer. We’re going to need a lot of coders who can turn complex business needs into quantum algorithms and code, and they won’t need to have Ph.D.’s either. However, the software engineer who’s our guest today does happen to have a Ph.D. — in stellar astrophysics. We’ll talk about how she went from working on the potential habitability of exoplanets to working on a software platform that can help companies with portfolio optimization, derivative pricing and other use cases — stellar topics after reaching a half-year milestone with this episode of The Post-Quantum World.

I’m your host, Konstantinos Karagiannis. I lead Quantum Computing Services at Protiviti, where we’re helping companies prepare for the benefits and threats of this exploding field. I hope you’ll join each episode as we explore the technology and business impacts of this post-quantum era. 

Our guest today is a quantum software engineer at Agnostiq. She’s rising in the field of quantum development, but she started out by reaching for the stars literally with a Ph.D. in stellar astrophysics. I’d like to welcome to the show Dr. Anna Hughes.

Hi, Konstantinos. Thank you so much for having me on.

Yes. Thanks for joining here. You had a unique past to quantum development, so, first, I’d love to hear a little bit about your work in astrophysics. For listeners, trust me, there’s a quantum computing reason I’m asking this that we’ll get to.

I finished my Ph.D. at the University of British Columbia this past summer, where I was studying these things called ultracool dwarfs, which are the smallest and coolest stars in the universe. I was looking for radio emission from these stars to try to learn something about their magnetic behavior, which we see the traces of in radio emission. I found that about 10% of these objects — from a pretty small sample, because these radio observations do require a lot of time on very high-powered telescopes — seem to have tracers of violent magnetic activity that could pose a threat to the planets that are orbiting around these stars if they do have them. That was pretty exciting, and I was trying to understand, because we do see a lot of planets around these low-mass stars, whether or not the stellar activity poses a threat to those planets.

In that role, you were basically doing the same thing, in that you had very limited access to machines because of their scarcity for study.

Yes, exactly. There were only a couple — at a certain frequency range, only one — that I could access that was powerful and sensitive enough to be able to measure the radio emission from these small stars.

You’re right back where you started now, with very limited access to a very few powerful machines?

Exactly, yes — very similar.

Not to get too off-topic, but when I heard about the stripping-wave-atmosphere stuff when I looked into your work, I just wanted to know if you had any thoughts on the Fermi paradox. For listeners, that’s this seeming paradox about extraterrestrial life — so, the Drake equation, things like that, predict that there might be all these habitable planets and life, and then the paradox says, “Why don’t we hear from them? Why don’t we see them?” I don’t know if you had any thoughts on that crazy topic.

Yes. The Fermi paradox is a hard one. There are all sorts of answers, ranging from “Maybe we’re the first generation of intelligent life, so this is something that’s quite rare,” to that other civilizations might not be particularly interested in contacting us, or if the stellar behavior of low-mass stars truly is quite violent across the board and this is where most terrestrial — Earth-like — planets are orbiting, then it might be that the planets that have the potential to host life aren’t actually able to support that life because of the stellar activity. We don’t know if any of these factors are really behind the Fermi paradox, but they definitely complicate the situation.

I wanted to throw something like that in and bring up this next point: Before we get to Agnostiq, how do you feel about using quantum computers to further science, like maybe what Feynman was thinking when he proposed in 1981 this idea that the only way to study the quantum entity that is the universe is to study it with a quantum system. Are you interested at all still in tying those worlds together?

Yes. That’s something that I hope that we start seeing down the line — a shift as we get more and more sophisticated in quantum devices, a shift from using only classical supercomputers for scientific research to maybe relying more on quantum devices.

I’d like to see that, too. I feel like there’s something there that we’re going to uncover. We’re still in early days. Who could predict all the use cases? Now, I’d like to see more scientific ones, definitely.

With that, let’s shift to some of what you’re doing now. Tell us a little bit about Agnostiq, the company.

We’re a team of scientists and software engineers that are working on these deeply technical problems. We’re trying to develop an ecosystem of tools and applications to make quantum computing more accessible to our clients, and we’re approaching this from a research perspective. A lot of us, like me, are coming in with Ph.D.’s in different areas of physics or in math, and that’s what we’re used to, and now we’re applying that to industry, where it’s similar, but you are moving at a faster pace. 

Essentially, what we’re doing is, as we enter this era of having more computing available and affordable to different companies, we want to build a bridge between those resources and between the companies that are interested in using them but maybe don’t know how. If you’re interested in using a quantum device but don’t have experience using quantum and don’t know how to use a quantum device, that’s OK. You can still access it through our algorithms.

So, there are the two sides: There’s building your platform, and then there is actually interacting with the customers. Is it the same teams interacting with both? Are you responsible for more interacting with customers or for building a platform?

My role primarily, so far, has been working on building the platform. I haven’t interacted with customers yet, but I also am quite new. I’ve been here for a couple months, so down the line, maybe I will do some of that.

So, you’re still building the core. I know you guys list three applications on the site: portfolio optimization, cyclical arbitrage and derivative pricing, and those are all obviously financial. Are these the ones that you feel already are canned quick proofs of concept that you can give to customers? Is that why they’re listed, because they’re already really well developed — you just plug in their data and get going?

Customers would have to contact us directly, because it is going to depend on the particular problem that they have, but, yes, we do have a handful of algorithms that are —we’ve used them with particular applications, and they’re functioning quite well, including those three and, recently, diversification.

If customers had a completely different problem they want to approach, then it would be a different approach? There would be a whole other process involved, developing what they want with a little more handholding?

Yes. If they have their own particular application that they would want to write, or if they would want to use one of our applications as a template for customizing their own, then they can do that. They could access our algorithms that we have online as is. They just have to reach out to us.

I saw that it mentions that you have the purpose-built ones and then algorithmic libraries. How intuitive is that, and how extensive is that? Are there a lot, or is it really robust —anything someone can think of right now, where they look and say, “Here’s an example,” or is it just like a handful? I was curious, because I think you have visibility to that.

Yes. We have quite a few things online — we have portfolio optimization. You can access quantum gate-based algorithms, quantum annealing algorithms, CPU and GPU algorithms. We have visualization, data acquisition and preprocessing. If you are familiar with finance and that’s something you work in, but you’re not familiar with quantum computing but you would like to use a quantum device, then you could through us. If you are familiar with quantum computing, then that’s fine, too — you would be able to interact a little bit more if you wanted with the quantum device, but if you don’t want to interact with it at all, that’s OK. You wouldn’t have to with our applications.

If you do want to get hands-on, is it a complete development environment — similar to, let’s say, Composer if you went to Qiskit, or what’s the interface for someone who does want to get down into the weeds?

It would be a Python library, so you need to have familiarity with Python, and so you could write your own application and the Python script, or you could use one of ours.

So, it’s a library. I wasn’t sure. A lot of customers have sensitive workloads that they still don’t entrust to the cloud. That’s in a pure classical realm, and this is always amazing to me: I’ll talk to a financial customer, and they’re like, “Yes, these we just do in-house — these runs, these end-of-day kind of jobs and things,” and that’s strange. I can’t imagine having everything in-house now. They’re concerned with moving quantum to the cloud, too, and obviously, that’s how we access these machines, because that’s where they are.

I saw that your company does obfuscation. Can you talk a little bit about that protocol and how you accomplish that?

SI don’t work on it specifically, but the quantum tools are quantum native, and they’re inspired by quantum homomorphic encryption, so the exact label that you’re talking about is quantum circuit obfuscation, which is fundamentally similar to a traditional code obfuscation, but for quantum circuits. We have a patented protocol and a pipeline of products that will be available as hardware scales up. You just have to stay tuned, because we’ll have more to come about that.

Actually, when you generate the wallet, you can select which hashing algorithm you want. I think the default is SHA-256. SHA-256 is considered to be quantum resistant/secure, I suppose, depending on who you talk to. Finding the pre-image on something that’s been SHA-256 is quite difficult.

That’s great, because at the end, obviously, the quantum data is meaningless. When you get to the quantum computer, it’s not like someone’s personal information or a credit card number is there.

Yes, I don’t think that’s the same.

Yes, and that’s not there anyway, but this is more about protecting the approach — the IP that might be generated and how you go about doing something. You’re obfuscating the circuit and what some company might claim as their advantage or their edge. Advantage is a loaded word, of course.

Yes.

You mentioned you don’t work on the obfuscation, so what piece of the development puzzle do you work on, specifically?

So far, again, I’m quite new. I’ve been working on helping develop these algorithms as well as the applications and writing a little bit in the way of tutorials that we’ll have available as blog posts on how to use our applications, so we’re hoping that these applications will be a jumping-off point. These are Python scripts for our clients for if they want to do, for example, diversification using one of our algorithms — then it’ll be intuitive for them to just follow along and see how we’ve done it and then take that and go.

This is your first time, then, working with algorithms in a quantum computer? I was wondering how you made that jump, and what like inspired you to move over?

It is new in a lot of ways, and it’s the same in a lot of ways as well. Doing my Ph.D. in astrophysics, I was coding to solve very difficult and very technical problems as a research problem, and that’s not really that different from what I’m doing now. Doing my Ph.D. in Vancouver, there were a couple of quantum computing companies around — 1QBit is there, and D-Wave is there. As I was going through my Ph.D., quantum computing was something that I was always very aware of, and I’ve even seen friends go on to get their Ph.D.’s in astronomy and make that switch to working in quantum computing. That was a path that I definitely had always seen.

When you were making that jump, you realized that you had that ability to translate that problem-solving into the quantum realm, and that’s one thing we’re finding: that some people have a struggle with — a normal developer, they sometimes can’t go from a complex business idea and then bring it all the way to qubits. They sometimes struggle with that. We look for that. We look for linear algebra skills. We look for all these things in developers. What kinds of advice would you give to people who are trying to start out in quantum programming?

My advice would depend on what your background is, so, having done a Ph.D. in astrophysics, I did take graduate-level quantum mechanics, I had taken linear algebra, so if you’re coming from a background like that, where you do have some experience, then I would say it’s not as difficult to make that leap. But if you are coming from the computer end of things and aren’t as familiar with quantum, there are lots of tools online that I would recommend doing to familiarize yourself with the basics of quantum mechanics. I know PennyLane and Qiskit have these great tutorials that you can go through to learn how to build quantum circuits and things like this.

We found it’s not always easy for some people in the pure programming world to make that jump and understand the concepts right away, so it’s a unique skill having all these things at once. It’s like a perfect storm of goodness. 

Right now, your platform can let you compare the results of what you accomplish — across different hardware types, for example Do you have sophisticated reporting or visualization tools that customers can use?

We do have visualization. I don’t work on it specifically, so I’m going to have a limited comment on that, but if you wanted to see how your particular application works with a bunch of different devices — because our hardware in Agnostiq, it’s really easy to switch in and out different devices, and you can tune on your own time. If you want to see what device is going to work optimally for this problem, you can piece together which devices you want to use and which subroutines. This is something that we would encourage our clients to experiment with until they get that perfect combination of devices.

Because you’re helping create the platform, have you still been able to do any internal research where you still try to apply to a use case internally to test the code — that kind of thing?

Yes. That’s a large part of what I do: We’ll have the algorithm developed, and then we’ll think of a couple different ways in which it can be applied. The diversification use for our selector algorithm is a great example, where our selector algorithm will pick out the most dissimilar signals in a selection of signals that are also representative of their cluster.

That sounds very abstract, but the application of this would be if you’re trying to build a diverse profile and you have a bunch of different tickers and you’re saying, “What are the most dissimilar tickers — maximally diverse, but also still representative of their particular sector?” If you take those tickers and say you take their daily returns over the past year, now you’re working with time-series data and feed that into the selector algorithm. That can return for you a selection of however many diversified tickers that you would want. Something like that was pretty cool — putting together a time-series case for this particular algorithm that we had.

Do you then benchmark that against classical approaches to see how close you are to gaining some kind of — or how close you are to extrapolating some kind of —advantage for quantum?

Yes, that’s something that we’re actively working on.

Do you have any guesses, ballpark, to make about how that’s looking if you were to draw a graph off into space? This always fascinates me — how close people are getting in any particular application.

I’m going to do the whole toeing the line here. When quantum advantage will come, we’re seeing some pretty dramatic improvements in these small cases, and it’s quite promising, but I don’t have an answer for you about when exactly we’ll start to release a quantum advantage across the board.

I have a theory that annealers will give us our first real advantage with optimization. Do you agree?

I think quantum computers — especially annealing devices — are quite good at optimization problems. If you have a particularly specific optimization problem like you would get in a lot of finance cases, then, yes, you can start to see quantum computers are quite excellent at those.

When we talked to Sam Mugel at Multiverse, they were doing portfolio optimization where they’ve got some impressive numbers by comparing to classical. They were able to do, instead of 33 hours of a classical tensor-networks run, they were able to do it in three minutes, which is quite a big improvement, but they sacrificed some accuracy — about 20% accuracy.

Yes, that is the thing. A quantum device would be fantastic if you want to send an approximate answer, but you want it really quickly. If you want a very precise answer and you have a lot of time on your hands, then, yes, you’re going to want to go with a classical device.

Yes, for now. Are you doing any work on improving accuracy? Have you reached that level of granularity yet where you could start tweaking and seeing improvements to accuracy? On the code side. Obviously, hardware can get better. We all know that, but — it’s an abstract question.

It’s a hard question to answer. It very much depends on how you ask the question — the answer that you’re going to get — so if you’re very careful and you’re very precise about how you write your particular application or your particular algorithm, then you can edge closer to a more accurate answer, but that’s more on the algorithm end.

Do you and the team, do you have any benchmarks that you traditionally run now? Have you established the suite that you go to, to see how performance is doing with any given application?

That’s not something that I work on specifically, but I do know that it’s part of our consideration, too.

Yes. I have a feeling that there’s a lot more we could squeeze out of these machines with the software stack than we’re currently doing. We can still refine and get better results and better approach at combining our answers, too, and running them on different machines and comparing them. I’m always curious if anyone has come up with any creative ways I haven’t thought of in that space.

Do you have any favorite pet projects you want to apply quantum to when you’re ready at the company? Is there something that’s been burning in you, like, “I just can’t wait to apply to this?”

When I think about particular examples of how you can apply quantum algorithms to new applications, even thinking outside of finance, which is our main focus with the Agnostiq finance library, having this background in astrophysics, a lot of times, astronomy questions will occur to me. For example, if we’re finding a way to address outlier detection and we’re thinking about doing outlier detection in light curves, I think, “I have all of these solar light curves on my computer. We could break that up into different chunks, and we could see when the sun is having heightened periods of activity as an example of outlier detection so that you could pick out at this particular time chunk — the sun’s activity became dissimilar to these other time chunks.”

That’s very interesting. I feel like quantum now almost gives you an instant new realm of papers that can be written. There are papers, any number of papers, written on any subject, and then you could just say, “Now, let’s apply quantum. This is how we did it, and they’re fast.”

Exactly. It’s, “How do we turn this problem into a problem that would be solved with the quantum device?” and a lot of that is turning a problem into an optimization problem.

Or there’s the whole idea of looking at periodicity of numbers — like what, essentially, Fourier transformations were used for originally — and then, of course, they’re going to be applied to Shor’s algorithm. You can possibly be looking for numerous things from the distant data that were gathered. Just by applying it to quantum computing, it becomes a data problem all of a sudden. Have you done a lot with machine learning and anything in that space?

Yes. That’s something that we’re definitely dipping our toes into in quantum machine learning, and there are a couple different things with PCA that I’ve done on my end that has been pretty exciting. PCA is principal component analysis, so that’s something that I’ve written a tutorial on that we’ll have as a blog post later, but, yes, there are a few applications.

I find that to be this extra-fascinating layer now. No matter what science you’re looking at — that’s why I asked that question earlier about how Feynman thought about quantum computing. It just feels like, “Yes, we read all those papers, and now let’s layer on quantum.” What else can we find? What else is hiding in this data, in the huge data sets that we’ve gathered over the years in whatever discipline? I’m definitely excited about more of that. 

How do you feel about the state of sharing information in quantum information science right now, and do you feel like it’s still open enough? Do you feel like companies are going dark? I ask guests this question regularly: Do you feel like people are going hush-hush already and trying to generate IP to the detriment of society?

I haven’t seen a problem with that so far. Every time that I have something new that I need to learn — which is quite a lot, since I am new to this field — I found the information readily available, but it’s probably the case that as you start getting into deeper and more specific problems, companies are going to start not wanting to share as much as something out of a university.

Has anyone at the company said, “Whoa, this customer came in, and they had this idea that we never thought of, and it blew our minds?” Have you heard anything like that in the virtual hallways, or is it still that your team is ahead of the curve?

I haven’t heard that particular comment yet, no. It still seems to be that the new ideas are coming internally.

It’s still so rarely that a customer dazzles you and you’re like, “Whoa — wow! Who would’ve thought of such a thing?” It’s still pretty rare. We do workshops that are big groups of customers, and every once in a while, someone will post one in the virtual environment that we create for them, and we’re like, “Wow, that’s actually a good idea. Why did you just share that with five competitors? I don’t know why you did that.” But normally, it’s tried-and-true things. 

If you keep applying and working on papers in your other field, you’re not really letting that go. It doesn’t sound like you’re truly letting go of the astrophysics side.

I don’t plan on continuing to do astrophysics-specific research. There are a few small projects that I can think of that are very easily done in the quantum world, and I think that it’s a fun way to show that this particular quantum algorithm works. The focus in that case is the quantum algorithm and not the astrophysics. It’s like, “Here’s a fun little thing that we used the algorithm on,” but the astrophysics itself is not going to be the focus.

It sounds like you are a true convert then.

I am a true convert, yes.

You’ve given up zeros and ones forever in the binary world. The future papers you see publishing, then, you expect they’re going to be heavily on the quantum focus and possibly having nothing to do with astrophysics at all, right?

Yes. We’ve just started working on one which is pretty exciting because it will be my first, but yes, it’s a quantum-centered paper, not an astronomy paper.

Just a little bit about your company’s culture — is it like Google? Is there that 20% time or whatever number — they change it all the time. Do you have that policy there where there’s a certain amount of time they want you, and encourage you, to do research that would just lead to publication?

Things are quite free, especially on the research end, because we do want to have a research approach to these, so we’re welcome to take our own initiative in that sense. There’s no hard line of, “This many minutes a day is what you may spend doing your own research.”

That’s great. That’s the approach I take over here, too. I was talking to one of my employees about a paper. She wants to write. I try to encourage that. I think it’s great, and it could lead to other actual problem-solving steps that the company could benefit from.

Yes. There are all kinds of things that occur to you as you’re starting in a research project: “We could definitely use a quantum computer for this, and this could definitely be framed as an optimization problem.”

Yes, I agree. It sounds like you’re off to a great start with this new frontier here, so I hope to read some of your work in the very near future. Thanks a lot for coming on. I really appreciate it.

Thank you for inviting me on.

That does it for this episode. Thanks to Dr. Anna Hughes for joining today to discuss her work at Agnostiq, and thank you for listening. If you enjoyed the show, please subscribe to Protiviti’s The Post-Quantum World and leave a review to help others find us. Be sure to follow me on Twitter and Instagram at KonstantHacker. You’ll find links there to what we’re doing in Quantum Computing Services at Protiviti. You can also find information on our quantum services at www.protiviti.com, or follow Protiviti Tech on Twitter and LinkedIn. 

Until next time, be kind, and stay quantum curious.