Karen Christensen

In September of 2019, Peter Wittek, an assistant professor at the ��߲ݴ�ý, went missing during a mountaineering expedition in the Himalayas after reportedly being caught in an avalanche. A search and rescue mission was launched but the conditions were very difficult and Wittek was not found.

“Peter’s loss is keenly felt,” said Professor Ken Corts, acting dean of the Rotman School of Management. “He was the Founding Academic Director of the CDL Quantum Stream, a valued instructor in the MMA program, data scientist in residence with the TD Management Data and Analytics Lab, an exceptional contributor to Rotman and U of T – and a wonderful colleague.”

A ceremony to remember Wittek will take place on Feb. 3 from 3 to 4:30 pm in Desautels Hall at the Rotman School of Management.

Quantum computing and quantum machine learning – an emerging field that counted Wittek as one of its few experts – was the topic of his final interview in Rotman Management Magazine. It is reprinted below:

You oversee the Creative Destruction Lab’s Quantum stream, which seeks entrepreneurs pursuing commercial opportunities at the intersection of quantum computing and machine learning. What do those opportunities look like?

We’ve been running this stream for three years now, and we were definitely the first to do this in an organized way. However, the focus has shifted slightly. We are now interested in looking at any application of quantum computing.

These are still very early days for quantum computing. To give you a sense of where we are at, some people say it’s like the state of digital computing in the 1950s, but I’d say it’s more like the 1930s. We don’t even agree yet on what the architecture should look like and, as a result, we are very limited with respect to the kind of applications we can build.

As a result, focusing on quantum is still quite risky. Nevertheless, so far we have had 45 companies complete our program. Not all of them survived, but a good dozen of them have raised funding. If you look at the general survival rate for AI start-ups, our record is roughly the same – and given how new this technology is, that is pretty amazing.

What are the successful start-ups doing? Can you give an example of the type of problems they’re looking to solve?

At the moment I would say the main application areas are logistics and supply chain. Another promising area is life sciences, where all sorts of things can be optimized with this technology. For instance, one of our companies, Protein-Qure, is folding proteins with quantum computers.

Finance is another attractive area for these applications. In the last cohort we had a company that figured out a small niche problem where they had both the data and the expertise to provide something new and innovative; they are in the process of raising money right now. The other area where quantum makes a lot of sense is in material discovery. The reason we ever even thought of building these computers was to understand quantum materials, back in the 1980s. Today, one of our companies is figuring out how to discover new materials using quantum processing units instead of traditional supercomputers.

We have a company called Agnostic, which is doing encryption and obfuscation for quantum computers. Right now IBM, Rigetti Computing and D-Wave Systems are building quantum computers for individual users. They have access to everything that you do on the computer and can see all the data that you’re sending. But if you’re building a commercial application, obviously you will want to hide that. Agnostic addresses this problem by obfuscating the code you are running. One application we’ve seen in the life sciences is a company called EigenMed, which addresses primary care. They provide novel machine learning algorithms for primary care by using quantum-enhanced sampling algorithms.

We also seed companies that don’t end up using quantum computing. They might try out a bunch of things and discover that it doesn’t work for the application they have in mind, and they end up being 100 per cent “classical.” StratumAI is an example of this. It uses machine learning to map out the distribution of ore bodies under the ground. The mining industry is completely underserved by technology, and this company figured out that to beat the state-of-the-art by a significant margin, it didn’t even need quantum. They just used classical machine learning – and they already have million dollar contracts.

Which industries will be most affected by this technology?

Life sciences will be huge because, as indicated, it often has complex networks and probability distributions, and these are very difficult to analyze with classical computers. The way quantum computers work, this seems to be a very good fit, so that is where I expect the first killer app to come from. One company, Entropica Labs, is looking at various interactions of several genomes to identify how the combined effects cause certain types of disease. This is exactly the sort of problem that is a great fit for a quantum computer.

You touched on quantum applications in primary care. If I walked into a doctor’s office, how would that affect me?

It’s tricky because, like mining, primary care is vastly underserved by technology. So, if you were to use any machine learning, you would only do better. But EigenMed was actually founded by an MD. He realized that there are certain machine learning methods that we don’t use simply because their computational requirements are too high – but that they happen to be a very good fit for primary care, because the questions you can ask the computer are similar to what a GP would ask.

For instance, if a patient walks in with a bunch of symptoms, you can ask, “What is the most likely disease?” and “What are the most likely other symptoms that I should verify to make sure it is what I suspect?” These are the kinds of probabilistic questions that are hard to ask on current neural network architectures, but they are exactly the kind of questions that probabilistic graphical models handle well.

Are physicians and other health-care providers open to embracing this technology, or do they feel threatened by it?

First of all, health care is a heavily regulated market, so you need approval for everything. That’s not always easy to get and, as a result, it can be very difficult to obtain data. This is the same problem that any machine learning company faces. Fine, they have this excellent piece of technology and they’ve mastered it, but if you don’t have any good data, you don’t have a company. I see that as the biggest obstacle to machine learning-based progress in health care and life sciences.

You have said that QML has the potential to bring about “the next wave of technology shock.” Any predictions as to what that might look like?

I think it’s going to be similar to what happened with deep learning. The academic breakthrough happened about nine years ago, but it took a long time to get into the public discussion. This is currently happening with AI – which, at its core, is actually just very simple pattern recognition. It’s almost embarrassing how simplistic AI is – and yet it is already changing entire industries.

Quantum is next – not just quantum machine learning but quantum computing in general. Breakthroughs are happening every day, both on the hardware side and in the kind of algorithms you can build with quantum computers. But it’s going to take another 10 years until it gets into public discussions and starts to disrupt industries. The companies we are seeding today are going to be the ones that eventually disrupt industries.

Alibaba is one of the companies at the forefront of embracing quantum, having already committed $15 billion to it. What is Alibaba after?

First of all, I want to say a huge thank you to Alibaba because the moment it made that commitment, everyone woke up and said, “Hey, look: the Chinese are getting into quantum computing!” Almost immediately, the U.S. government allocated $1.3 billion to invest in and develop quantum computers, and a new initiative is also coming together in Canada.

The world’s oldest commercial quantum computing company is actually from Canada: D-Wave Systems started in 1999 in British Columbia. Over its 20-year history, it managed to raise over $200 million. Then Alibaba came along and announced it was committing $15 billion to quantum and this completely changed the mindset. People suddenly recognized that there’s a lot of potential in this area.

What does Alibaba want from quantum? You could ask the same question of Google, which is also building a quantum computer. For them, it’s because they want to make their search and advertisement placement even better than it already is. Eventually, this will be integrated into their core business. I think Alibaba is looking to do something similar. As indicated, one of the main application areas for quantum is logistics and supply chain. Alibaba has a lot more traffic than Amazon. Its orders are smaller, but the volume of goods going through its warehouses is actually much larger. Any kind of improved optimization it can achieve will translate into millions of dollars in savings. My bet is that Alibaba’s use of quantum will be applied to something that is critical to its core operation.

The mission of CDL’s Quantum stream is that, by 2022, it will have produced more revenue-generating quantum software companies than the rest of the world combined. What is the biggest challenge you face in making that a reality?

People are really waking up to all of this. There is already a venture capital firm that focuses exclusively on quantum technologies. So, the competition is steep, but we are definitely leading in terms of the number of companies created. In Canada, the investment community is a bit slow to put money into these ventures. But every year we are recruiting better and better people and the cohorts are more and more focused – and, as a result, I think we are going to see more and more success stories.

It seems like everyone is interested in quantum and they are thinking about investing in it, but they are all waiting for somebody else to make the first move. I’m waiting for that barrier to break and, in the meantime, we are making progress. Xanadu just raised $32 million in Series A financing, which indicates that it has shown progress in building its business model and demonstrated the potential to grow and generate revenue. ProteinQure raised a seed of around $4 million dollars. And another company, BlackBrane, raised $2 million. So, already, there are some very decent financing rounds happening around quantum. It will take lots of hard work, but I believe we will reach our goal.

Peter Wittek was an Assistant Professor at the Rotman School of Management and Founding Academic Director of the Creative Destruction Lab’s Quantum stream. The author of Quantum Machine Learning: What Quantum Computing Means to Data Mining (Academic Press, 2016), he was also a Faculty Affiliate at the Vector Institute for Artificial Intelligence and the Perimeter Institute for Theoretical Physics.

This article appeared in the Winter 2020 issue of Rotman Management Magazine. Published by the ��߲ݴ�ý’s Rotman School of Management, Rotman Management explores themes of interest to leaders, innovators and entrepreneurs.

'We will have raised Canada’s game': U of T's Ajay Agrawal on the Creative Destruction Lab's past, present and future

Christopher.Sorensen — Fri, 21 Dec 2018 15:17:31 +0000

'We will have raised Canada’s game': U of T's Ajay Agrawal on the Creative Destruction Lab's past, present and future

Christopher.Sorensen Fri, 12/21/2018 - 10:17

Cutline

Ajay Agrawal, the founder of the Creative Destruction Lab, is the Geoffrey Taber Chair in Entrepreneurship and Innovation and a professor of strategic management at U of T's Rotman School of Management (photo by Yana Kaz)

Karen Christensen

Topic

Global Lens

Artificial Intelligence

Computer Science

Creative Destruction Lab

Global

Graduate Students

Innovation & Entrepreneurship

Research & Innovation

Rotman School of Management

Startups

The Creative Destruction Lab – a seed-stage accelerator affiliated with the ��߲ݴ�ý's Rotman School of Management – is a startup success story in its own right.

Since its founding in 2012, CDL has undergone a massive expansion, boasting multiple streams of startups and partnerships with business schools across Canada and into the United States.

It has also emerged as an early leader when it comes to supporting startups in the field of artificial intelligence – technologies many believe are poised to revolutionize everything from transportation to medicine.

“To our knowledge, the CDL is home to the greatest concentration of AI-based companies of any program on Earth,” says CDL founder and Rotman Professor of Strategic Management Ajay Agrawal.

In an interview with Rotman's Karen Christensen that originally appeared in the winter issue of Rotman Management, Agrawal discusses the origins of CDL's focus on AI, how the lab benefits business school students and why it decided to make a “bold” move into the quantum realm.

Note: This interview has been edited and condensed. The full interview can be accessed here.

Three years ago, CDL made a huge bet on artificial intelligence and machine learning. What prompted that?

In our first year of operation, one of the startups that came to us was Chematria, now called Atomwise. Its founder, Abe Heifets – a U of T PhD in computer science and biology – was applying a new AI technique to drug discovery. What Abe was doing represented not just a marginal improvement, but a potentially transformative change to the way drugs are discovered, which represents a multibillion-dollar problem for the pharmaceutical industry.

While we were working with Abe, a team of graduate students from U of T computer science won a high-profile competition at Stanford called ImageNet. It’s basically an image-recognition competition, whereby a computer is given a bunch of pictures and has to identify the image, whether it’s a ball, a horse or a wheelbarrow. This team from Toronto participated, and not only did they win – using a machine learning technique called deep learning, largely developed at U of T – but they won by such a margin that the following year, all of the finalist teams were using their technique.

Those are just two examples of events that inspired us to bet on machine intelligence. Overall, we saw mounting evidence that AI was a general-purpose technology that could be applied to a wide range of problems across a vast array of industries, and that’s what prompted us to dedicate a new stream of the Lab’s activity to AI.

At first, you faced resistance. Why?

People said we were being too narrow, that there weren’t enough startups to fill an AI stream and that there wasn’t enough interest from investors. At the same time, we had believers. One such believer who herself had written a highly influential blog post describing the “landscape” of companies emerging in the machine learning world was Shivon Zilis – a Canadian based in San Francisco and a partner at the venture investing firm Bloomberg Beta, where she led the firm’s investments in machine intelligence. I invited her to the Rotman School to present her insightful analysis to our MBA students, and the CDL team – quickly realizing she is a star – recruited her to join forces on our AI initiatives. Elon Musk subsequently saw the same potential and recruited her to help him build his empire.

So, we moved forward with the new stream, but to address these concerns, in 2015 we also launched an annual conference – with Shivon as co-chair – called Machine Learning and the Market for Intelligence. The goal was to educate the Canadian business community about the importance of this emerging field. Leaders in the field – from organizations like Google, Uber, Apple, Stanford, Carnegie Mellon and MIT – came to Rotman to discuss and debate how AI is and will impact a variety of fields, from life sciences to manufacturing to retail. We held our third annual conference in October 2017.

Talk a bit about the CDL’s results to date.

The launch of our AI stream transformed the lab from a Canadian enterprise into a global one. In our first year, our startups were all from Ontario, but they now come from around the world. Similarly, in our first year, our fellows were all from Canada, and that, too, changed when we launched the AI stream. Our ML7 – Machine Learning Seven – includes William Tunstall-Pedoe, who flies in from Cambridge, England, every eight weeks. He has a PhD in machine learning and founded Evi, which was acquired by Amazon in 2012. Evi’s technology powers the AI engine in Amazon’s Alexa, which, to my knowledge, is still the top-selling consumer AI hardware product in the world.

The ML7 also includes Barney Pell, who flies in every eight weeks from San Francisco. Barney also has a PhD in machine learning and led an 85-person team at NASA that flew the first AI into deep space. He then built an AI company called Powerset that was acquired by Microsoft, and now he’s the co-founder of Moon Express, which is essentially building a Federal Express-type service to the moon, because Barney believes the moon is going to be an important gateway for commercial space travel.

So far, the results have surpassed our expectations. Back in 2012, we accepted 25 companies into our general high-tech stream. Last year, we doubled that by adding the second cohort focused on AI, so we had 50 startups. This year, we doubled our intake again by accepting 100 AI-focused startups and adding a new stream: The world’s first program focused on launching startups predicated on quantum machine learning (QML). To our knowledge, the CDL is home to the greatest concentration of AI-based companies of any program on Earth.

The Lab is one of the most popular second-year MBA courses at the Rotman School. Why does it resonate so much with students?

For two reasons: First, it combines the traditional mode of learning from lectures with learning-by-doing; and second, it links academic work with a sense of ownership. The traditional approach to learning at CDL is led by our chief economist, Professor Joshua Gans, who developed a structure for teaching entrepreneurial strategy along with MIT’s Scott Stern. This provides students with an academic framework and context for what they’re going to experience next. Then comes the learning-by-doing part. Normally, business schools use Harvard Business School cases to provide examples in the classroom. We replaced those with real companies. Working with founders, fellows and associates provides students with an opportunity to roll up their sleeves. Instead of reading a 30-page case that comes with a fact set, they have to find the facts themselves and figure out – of the infinite information out there – which bits are the most valuable for their needs. They experience the messiness of the real world and the reality of having to make decisions without having full information.

The second piece is ownership. When our students work with these startups, every decision matters, so they have a real sense of ownership. It’s a powerful learning experience to feel ownership over the results because the consequences are so tangible.

Universities rarely adopt programs developed elsewhere. What motivated UBC, NYU and others to adopt the Creative Destruction Lab program?

Every university has a program or course on entrepreneurship and startups, but I think the CDL stands out due to its significant results. The calibre of investors from the business community who have rallied around the CDL is unprecedented. Naturally, other universities would love for that to happen at their own business schools.

When UBC indicated interest in adopting the program, the big question was, ‘Is this replicable?’ But a very competent team, under the direction of UBC Professor Paul Cubbon, was able to reproduce it. When CDL-West completed its first year, the results on all dimensions were impressive, and we had evidence that, yes, this program is replicable. We have since launched CDL at the University of Calgary, Dalhousie University and Université de Montreal, and in October, we announced a partnership with New York University’s Stern School of Business.

CDL Toronto’s competition is not Vancouver, Calgary, Montreal, New York or Atlantic Canada: It’s Silicon Valley. Each of the CDLs has attracted some of the top business people from its region. Our challenge now is to cross-pollinate, so that the Montreal fellows are connecting with companies in the Toronto program and the Toronto fellows are connecting with companies at CDL Atlantic, and so on. One of the things that makes the Bay Area so effective is that everything moves so fast. If we can accelerate the velocity of business development here, we will have raised Canada’s game as a whole.

You mentioned earlier that CDL launched the world’s first program focused on quantum machine learning. What is your vision for this initiative?

It’s a bold one: By 2022, the QML Initiative will have produced more well-capitalized, revenue-generating quantum machine learning-based software companies than the rest of the world combined, with the majority based in Canada.

Why QML? First, we can leverage the leadership that CDL currently has in the commercial application of machine learning. Second, we can leverage Canada’s leadership in quantum computing at places like the Perimeter Institute and the Institute for Quantum Computing in Waterloo, Université de Sherbrooke in Quebec, and D-Wave in Vancouver, among others. Third, we can leverage the network of investors, entrepreneurs, scientists, and corporations that have rallied around the CDL and our mission of commercializing science for the benefit of humankind.

Clearly, the CDL is leading the way in this arena.

I believe so. Three years ago, it felt like we were moving early on AI, but we realize now that – if we could turn back the clock – we actually should have started even sooner and moved faster. We were roughly a year ahead of everyone else, but now a number of programs in other countries are focused on AI startups – so we’re running fast just to keep our position.

In terms of QML, so far we’re the only ones doing it, and that’s because the technology is so embryonic. We might go for two or three years without a significant success, because we
might be too early. The point is, once there’s a hit, places like MIT, Stanford and Silicon Valley will all double down in this field. Our approach is to get ahead, make the investments now, and attract all the elements of the ecosystem to Canada.

We basically want to do in Toronto with QML what Silicon Valley did with semiconductors in the 1960s. There’s nothing inherently magical about Silicon Valley. The semiconductor industry happened to start there due to the pioneering efforts of a handful of people, and once that community grew big enough, it became very hard for other regions to compete. Our view is, if we can seed it here and if the industry takes off five years from now, by that time, Canada will have such a critical mass that it will be hard for the whole community to move somewhere else. We’re trying to plant the seeds now.

Already, three top Silicon Valley venture capitalists are sufficiently optimistic about this program that they offered to invest in every one of the companies that gets into it – sight unseen. Most of these companies won’t make it – and they know that – but they want to be involved because along the way, they will get an education in QML, and there is some positive probability that one or two of these companies will figure out a commercial application.