Sue Mniszewski has been a research staff member at Los Alamos National Lab (LANL) for over forty years and in that time has watched several novel architectures come and go.
While it’s but one of several revolutionary trends in computing over the decades, quantum computing certainly has her attention, especially for forward-looking graph partitioning and optimization problems at exascale.
LANL was one of the first sites to install a D-Wave quantum annealing machine and also has access to other quantum offerings through the various cloud interfaces of IBM and others. But for Mniszewski, the hardware is not the interesting part of the conversation just now, it’s seeing just how quickly she sped through toy problems on early devices—blowing past the capabilities of current limited-qubit machines.
Her focus now is seeing what next-generation quantum-classical applications might look like. To do that means simulating quantum capabilities of next-generation applications and while it doesn’t take exotic quantum systems on-prem to do that, it does take quite a bit of algorithmic work.
“It is these problems that are too big to fit on existing quantum machines that require quantum plus classical approaches where you need a classical solver working with a quantum device to solve the problem,” Mniszewski explains. To do this at functional scale, LANL recently struck up a three-year agreement with Quantum Computing, Inc. (QCI), which listed on NASDAQ today (QUBT).
At the core of this work are the kinds of graph partitioning and optimization problems that have a role in national security applications, among others. To get a handle on how quantum and classical computing techniques might mesh for this, Mniszewski and team are using QCI’s Qatalyst platform, which will use qubits and CPUs to more optimally partition large graphs as well as convert graphs into a form that can be handled on LANL’s D-Wave annealer with an eye on gate-based systems in the future. “We would like to see how many large problems we can solve this way, along with looking at other combinatorial optimization problems,” she adds.
LANL classifies QCI as experimental middleware since it’s not necessarily at the applications level and is not a hardware-specific platform, either. In fact, part of what makes it interesting to labs like LANL is that QCI is able to work across different devices—and varying qubit types.
“This is an opportunity to look at algorithms in a different way, to enable them using quantum mechanical effects. It’s about turning a problem upside down. And in some cases, when we compare these graph algorithms run classically versus quantum, we’re seeing better or comparable results, at least for the graph and optimization problems we’re running,” Mniszewski says. “There’s something here. We’d like these quantum/classical approaches combined with existing HPC systems.”
As for QCI, it’s a little difficult to talk about what they do for the same reasons it took some thought to classify its role at LANL. Middleware works as a term but it’s also something of a quantum simulator, even if QCI tries to avoid that term.
“We don’t use the term simulator because in some contexts it means an accurate embodiment of a quantum processor. That’s not what we do. It’s a classical computing resource using our approach to solve constrained optimization problems,” says Robert Liscouski, CEO of QCI.
What the core of the platform does is to allow quantum methods to be performed on classical systems without a lot of software complexity. It allows analysts and programmers to run the same program across any quantum or classical machine with no programming required. “Qatalyst also uses the same quantum-ready techniques to solve computational problems on classical computers, accelerating and delivering a diversity of highly accurate results. As a result, QCI is making the value of quantum computing available to a much wider enterprise audience, to solve real business problems right now,” Liscouski says.
The quantum CEO knows his market too. He has held public sector roles in Homeland Security, the U.S. Department of State, and the Intelligence Science Board, which supplied guidance to the CIA and other intelligence agencies. The point is, Liscouski well understands several agencies are on the lookout for more scalable ways to tackle large graphs—a key tool in establishing hard-to-find connections between individuals and entities. The market opportunity in the public sector likely was not lost on him when the company was founded in 2018.
QCI Michael Booth is also no strange to large-scale government HPC/scientific computing. He’s worked at Cray Research, SGI, and more recently, in D-Wave’s benchmarking division. VP of Product Development, Steve Reinhardt has a similar path over his forty years in high-end computing, also serving time at SGI and Cray Research before moving to D-Wave. His emphasis since 2003 has been on graph analytics and its role in cybersecurity, which helps explain his time at the Cray spin-off “YarcData” (thank god we haven’t had to write that word for years) working on graph analytics problems.
In other words, QCI is well-connected in the realms of national lab-class supercomputing, a big help at this early stage of quantum computing where the first buyers of quantum hardware will be (and have been) and the leaders in making quantum interplay with traditional HPC applications.
“Quantum computers demand immense effort to transform from classical problems into complex quantum programs. The tools required for this, quantum software development kits (SDKs), are extremely expensive and complicated, requiring quantum experts to write even simple programs, which can take months or years to successfully run. Additionally, each problem submitted to quantum computers must be updated to match the proprietary requirements of different machine hardware, reprogrammed with the same arduous, complex process and redeveloped with similarly lengthy timeframes,” Luscouski concludes.
With a listing on the NASDAQ, QCI might draw some attention from those outside of their known circles in supercomputing, But again, it’s early days. What would be really useful is to discover who might be interested in this quantum/classical approach at this juncture—there’s probably a graph result for that.