Catching up with Perimeter’s Quantum Intelligence Lab

In 2018, Roger Melko launched the Perimeter Institute Quantum Intelligence Lab (PIQuIL) to explore how Artificial Intelligence (AI) could be applied to problems in quantum physics. Seven years on, PIQuIL is going strong, with exciting new projects in quantum computing and talented alumni making waves across the quantum and AI industry.

What is PIQuIL and where did it come from?

In 2012, the development of AI passed a major milestone. AlexNet, a machine learning program created by Toronto-based researcher Geoffrey Hinton (of 2024 Nobel prize fame) and his students, managed a remarkable score in the ImageNet challenge. The goal of the challenge: identify distinct objects in an image. It’s an easy task for a human, but an enormously complex problem for a computer. AlexNet’s success showed that neural networks were leaving the realm of theory and becoming viable tools for application in many realms of data science.

It wasn’t long after this milestone that Melko came across neural networks himself for the first time. One of his PhD students, Bohdan Kulchytskyy, participated in internship with D-Wave, a quantum computing company based in British Columbia that was experimenting with these new networks. 

Melko soon realized that AI might have a place in computational physics. As a condensed matter physicist, Melko was studying how large groups of particles interact to create macroscopic behaviours, like superconductivity and phase transitions. In this line of research, as the number of particles grows, the physics becomes more complex – just as image recognition becomes harder as the number of pixels increases. Hinton’s AI had showed promise in the latter – could it also solve physics problems in the former?

Melko started to investigate and in 2017, published a highly influential paper with Perimeter postdoctoral researcher Juan Carrasquilla in Nature Physics that showed neural networks could indeed be helpful in studying the phases of matter.

This was still half a decade before AI hit the public eye with the advent of large language models (LLMs) in 2022. But Melko recognized its value to physics and sought to grow his research program.

So PIQuIL was born. Not only did it provide research space, it was also a vehicle for collaboration – between researchers at Perimeter Institute, the University of Waterloo, the Institute for Quantum Computing (IQC), and industry partners too. Vancouver-based 1QBit was the first of these industry partners, and the National Research Council of Canada also signed on.

Since then, PIQuIL has been dedicated to solving difficult physics problems using AI tools, in partnership with industry and the Canadian research community. 

Benjamin MacLellan, a current PIQuIL PhD student, believes the lab is something unique. “PIQuIL is this incredible intersection of scientific research and a nexus point for academia and industry,” he says. “Being in PIQuIL embeds you within a vibrant ecosystem, including connections to the local Waterloo physics and entrepreneurship communities, and with close ties to world-class global research institutes and companies.”

All of those industry and research connections, alongside the talent of researchers like MacLellan who pass through PIQuIL, enable the lab to make an outsized impact on the field. The results thus far have been impressive.

So what has PIQuIL achieved?

PIQuIL is now a leader in solving physics problems using AI tools. In recent years, it has grown beyond simulations for quantum matter and spilled over into a related problem: quantum computing. 

Some of PIQuIL’s early projects in this area involved pulling data off a quantum computer to create a digital twin of the device. This allowed researchers to iterate on the design of the quantum computer and improve it. Given that AI is a data-driven tool, it became clear they could use this same data to train new AI agents with unique capabilities.

From there, they began investigating how AI might help solve certain pressing challenges unique to quantum computing: quantum computers are very difficult to control and require complex error correction codes to keep them functioning.

PIQuIL’s approach to these problems grew into a ‘virtuous cycle’, improving both quantum computing techniques and AI tools side-by-side:

“We can use the data from quantum computers to train AI agents,” says Melko. “But we can also use AI agents to control quantum computers in a very technical sense – we have to manipulate qubits at a microscopic level.”

To do so, PIQuIL researchers often need to develop novel AI algorithms.

“These are very much inspired by neural networks, deep learning, and more recently, language models – we're using architectures that underlie LLMs. We adopt those strategies, but then we also develop them. Our AI tools are unique to PIQuIL. They're things we've developed in-house that we release open source out to the world,” says Melko.

A growing network of expertise

After seven years, PIQuIL has a significant network of students, postdocs, and alumni – experts trained to solve complex problems using AI – spread across the globe. A Slack channel of more than 130 experts keeps them all in touch, built up during the Covid-19 pandemic. It has become the heart of PIQuIL, where disparate researchers connect on difficult problems, as well as share career paths and technical advice.

“We have this massive, successful alumni network now,” says Melko. “The research we got into with PIQuIL – especially with that first generation of students when we were first movers – we were training physicists in cutting-edge research in artificial intelligence, and then they went out in the world and just slayed it. And they're still slaying it. They're paving the way for the current generations of students and postdocs that come out of the lab.” 

MacLellan, who joined a start-up even before coming to PIQuIL, is one of those students.  He found PIQuIL to be a perfect environment for bridging his two worlds: research and industry.

“The ability to drive my own intellectual property is particularly important to me. At PIQuIL, I can spend time working to bring research out of the lab, applying it to real-world problems, and commercializing new technologies. While working on my PhD, I've been deeply supported to continue building Ki3 Photonics Technologies, a spin-off from my past academic affiliation; I can take the ideas, inspiration, know-how, and support from PIQuIL and use them to drive forward the technological capabilities of our company,” he says.

Previous PIQuILers have found diverse outlets for their expertise after leaving. A few have stayed on at Perimeter in new roles, like Lauren Haward, who is now mentoring a whole new generation of graduate students as a teaching faculty member.

Some are in the AI world, like Anna Golubeva, who is now working at Zyphra Technologies, a Silicon Valley-based AI start-up. Others, like Giacomo Torlai, are working on quantum computing. As a PIQuILer, Torlai wrote one of the seminal papers on using machine learning for quantum error correction – a technique now used by companies like Google. Torlai currently works for Q-CTRL in California. 

Another on the quantum computing path is Kulchytskyy, who worked with 1Qbit as part of the partnership with PIQuIL. He’s now at Nord Quantique, a quantum computing company in Montreal.

A burgeoning ecosystem of start-ups

PIQuIL researchers have launch several start-ups of their own. One of these, the Canada-based yiyaniQ, provides quantum solutions to financial firms. It was founded by Perimeter research scientist Estelle Inack.

“PIQuIL is one of the very few research labs in the world that uniquely blends training, research, and entrepreneurship,” she says. “It is in this context that I can wear the hat of an academic entrepreneur – a position that allows me to simultaneously train graduate students in computational condensed matter physics, carry out world-class research at the intersection of condensed matter physics and machine learning, and go from lab to market by translating quantum intelligent algorithms to software products for the financial industry.”

Inack remains a research scientist at Perimeter Institute while carrying out her work with yiyaniQ, which has two other employees, and recently put out a call to hire a multi-year Mitacs postdoctoral position.

One of Inack’s PIQuIL colleagues, former postdoctoral researcher Aida Ahmadzadegan, launched a start-up in 2021 that used a combination of quantum algorithms and machine learning to help logistics companies optimize last-mile delivery services. Named ForeQast, it gained the support of the Creative Destruction Lab incubator program run out of the University of Toronto, and was one of nine companies to graduate from the program in 2022.

PIQuIL’s most recent start-up – a non-profit – grew out of an effort to make partnerships between industry and academia more productive, while at the same time pouring more resources into its quantum computing research program. Called Open Quantum Design (OQD), it is based around the trapped ion quantum computer at the University of Waterloo’s Institute for Quantum Computing. 

OQD takes what PIQuIL has learned about quantum computing control and artificial intelligence, and implements it into the control stack of the quantum computer. It is co-led by Melko alongside Crystal Senko and Rajibul Islam, who both head up research groups at IQC, and it runs independently as its own entity. OQD has a unique open-source charter, designed to make it easier for academic and industry partners to collaborate and make intellectual property barriers transparent. It’s a model pioneered in Silicon Valley: OQD is following in the footsteps of Linux, RISC-V, and PyTorch.

“Something that is open source serves as a solid root of trust,” says Melko. “It's completely transparent, and it's democratic.”

The track record PIQuIL researchers have bringing quantum and AI tools to bear on real-life problems is impressive, and their start-ups are evidence of their strong connections to the innovation ecosystem. The next few years will be an exciting time to watch them grow.

AI and Quantum Computing: a perfect match

In all of PIQuIL’s research efforts, from foundational research in quantum matter to the development of quantum computing, AI has proven to be a valuable tool for physics. 

Melko strongly believes it will be key to making quantum computing viable. What’s more, he and his team at PIQuIL are actively working to make it happen.

“I really think that you can't build quantum computers without a classical computer that controls it in all aspects, and it's become increasingly obvious that those classical devices need to run AI agents in order to really control quantum computers. They're just so difficult to control, and so a lot of our research effort now is along that line.”

Learn more about PIQuIL.