Silos may be good for grain, but they are not good for research, at least according to a cross-organizational team that brought together academics from the University of Bristol and Imperial College London and researchers from Hewlett Packard Labs and Hewlett Packard HPC Business Group.
This team wanted to find out if classical Gaussian boson sampling algorithms were as slow as other teams have maintained. They published their process and conclusions in a paper titled “The Boundary for Quantum Advantage in Gaussian Boson Sampling” (GBS) which has appeared in the prestigious scientific publication Science Advances. The paper details a ~10^9 speedup in efficiency over the previous state-of-the-art model.
As the authors note in their paper, “Identifying the boundary beyond which quantum machines provide a computational advantage over their classical counterparts is a crucial step in charting their usefulness.” GBS, in which photons are measured from a highly entangled Gaussian state, is a leading approach in pursuing quantum advantage.
The human league
The technology is important, obviously. But at least as important is the fact that it was sparked by human intuition and creativity. As project lead Thomas Van Vaerenbergh put it, you don't need to work in Labs to work on a Labs problem. There are ways to collaborate across innovation ecosystem to solve interesting problems.
“At HPE, we were looking for to engage further with our quantum computing,” says Van Vaerenbergh. “So I chatted to a couple of different people, including Bristol University, which is how I got to know Jake Bulmer.”
Subsequently, he also reached out to Diana Moise, performance engineer with HPE HPC, and Bryn Bell of Imperial College London, to put together a team that could take the question of what constitutes a quantum advantage and come to a conclusion that would push both quantum and classical computing forward.
“The question that started this project was, given that Bristol and Imperial have their algorithms, how can we at HPE help them do their research?” said Van Vaerenbergh. “I found there were some great HPE minds in Europe that were working in benchmarking teams and that they were indeed willing to have a look at that problem.”
So Labs had the photonics know-how, Moise had the hardware and programming skills, and the Bristol and Imperial teams had the quantum algorithms. Together, they were a sort of technological all-star team.
It wasn’t just the problem-solving capabilities that made this project useful. Customers have an abiding and deepening interest in quantum issues and HPE benefits from doing research in Europe, according to Van Vaerenbergh.
“This helps us make it clear that we're not just an American company, we're an international company and we do research everywhere,” he says.
Jake Bulmer, post-doc University of Bristol
One of the integral team members was Jake Bulmer, a PhD student at the University of Bristol, who was recommended to Van Vaerenbergh. This project was important enough for to Bulmer to include in his thesis. He was good enough to share his introduction to one of the chapters in which he talks about his experience with working across organizations for a common goal.
The project was started with a series of kick-off meetings, where I invited anyone who I thought would be interested in tackling the classical complexity of GBS. There were several good ideas immediately, but it quickly became apparent that Bryn Bell, who at the time was a Marie Curie research fellow at Imperial College London, had several unique insights which looked like they could lead to big gains and would be worth pursuing seriously.
At a similar time, Thomas Van Vaerenbergh, a research scientist at Hewlett Packard Enterprise (HPE) contacted my supervisor, Anthony Laing, as he was keen to learn more about how quantum researchers are using high-performance computing. Anthony put Thomas in contact with me, and after some discussion, Thomas agreed to put together a team of high-performance computing (HPC) software experts to help us scale our algorithms for running on HPC systems.
The theory and algorithm software development for this project was done in close collaboration with Bryn Bell. The scaling of the software using the Message Passing Interface (MPI) was done in close collaboration with Diana Moise, Allessandro Rigazzi and Jan Thorbecke of HPE. Analysis of the data we generated was carried out with help from Bryn, as well as Rachel Chadwick and Alex Jones.
“I would also add that myself, Bryn, and Alex are experimental physicists, and would normally be spending most of our time working on projects in the lab,” says Bulmer. “However, due to COVID, we had much more time for starting new projects involving theory and simulation.”
Diana Moise, performance engineer, Cray/HPE
“I think the goal for our collaboration was clear from the beginning,” says Moise. “It was understanding whether classical GBS algorithms are really as slow as another team claimed in their quantum experiment.
Moise’s part in the project was to parallelize the GBS algorithm that the Bristol team had developed and to run it on a large scale.
“But in the end, we got to achieve more than that, not only the scaling to ~128,000 cores, but also, thanks to the parallelization of the code, we were able to run much larger problem sizes than the earlier experiment. This had not been one of the original goals, as none of us knew if it was really going to work. But having several discussions as the work progressed, it became clear that the project was going to be successful and open up new possibilities.
“This was a collaboration spanning across different organizations, people with different backgrounds, and even countries,” says Moise. “But this is what made the collaboration very interesting and quite fun, to be honest. We all tried to bring to the project our know-how and really worked together to a common goal.”
The road to knowledge
It is arguable that such an exciting and fecund conclusion would never have been possible had the parties involved stayed in their own lanes. Instead, this “Wacky Races” approach leveraged everyone’s interest and distinctive skills to do something none of them would have been able to do alone. It also triggered a sense of fun and leveraged each researcher’s passion.
The paper was authored by Bulmer, Bell, Moise, and Van Vaerenbergh, along with Rachel S. Chadwick, Alex E. Jones, Alessandro Rigazzi, Jan Thorbecke, Utz-Uwe Haus, Raj B. Patel, Ian A. Walmsley, Anthony Laing
For more on the paper and the work behind it, read the press release, “Hewlett Packard Enterprise unveils supercomputing research that raises the bar for achieving quantum advantage.”
Curt Hopkins
Hewlett Packard Enterprise
twitter.com/hpe_labs
linkedin.com/showcase/hewlett-packard-labs/
labs.hpe.com
Curt_Hopkins
Managing Editor, Hewlett Packard Labs
