(upbeat music) - Hi, everyone, and thanks for coming back to Conversations at the Perimeter. Today, we're bringing you a conversation with Estelle Inack. She's a research scientist here at Perimeter, and she's also the Co-Founder and Chief Technology Officer of the company yiyaniQ. - I love this conversation with Estelle, partly because I found it a little challenging, the terminology like artificial intelligence and machine learning and neural networks.
These are terms that I've come across before in our work, but they get thrown around a lot in popular culture. And it was great to hear from an expert who's working, not just in these fields, but really finding the intersections between these fields. She was a very generous tour guide with us.
- I agree, and I really also loved hearing about how her work is really at the intersection of quantum science and artificial intelligence, but also at the intersection of academic research and industry applications. - And her personal story is pretty amazing too. You know, she's a scientist who's now working in a startup. She's trying to learn the business world. Estelle just has a fascinating personal story as well.
She's originally from Cameroon, and she originally wanted to do something completely different than physics. We won't give any spoilers, but her journey into physics was really fascinating, especially because she faced quite a lot of obstacles in her native Africa to becoming a physicist. And we learned that she's actually gone back to Africa to try to help inspire other women scientists there. - We're excited for you to hear the conversation. Let's step inside the Perimeter with Estelle Inack.
(upbeat music) Okay. Hi, Estelle. Thanks so much for sitting down with us today. - Thanks for the invitation. - It's great to have you here. - It's my pleasure. - So you work in a really exciting field that's also pretty new or at least rapidly growing, that is often called quantum intelligence. Can you tell us a little bit about what draws you to this field and why it's so exciting?
- So it's basically a very fancy name that means a lot of different things, depending on how you take different combinations of artificial intelligence and quantum computing. So for example, for some people it might mean using quantum computers to perform artificial intelligence tasks, in the field that they call quantum machine learning. For other people, it could mean using quantum computers with artificial intelligence for quantum control, for example,for quantum state preparation.
For other people, it can mean using machine learning techniques that you borrow from AI research to basically probe the behavior of quantum many-body system. And this is more of the field where I am now, borrowing machine learning techniques to probe the behavior of quantum many-body systems. - You mentioned a few terms there that I'm hoping you can elaborate on a bit. A lot of people have heard the term artificial intelligence. It's very much in the news.
I think a lot of people have heard the term quantum computing, maybe a little bit less so in the public consciousness. Can you tell us what those are and how you're sort of bridging the two fields? - Yes, so there are so many different ways, as I mentioned, and different ways of bridging the two phase. So quantum computers, for example, is just a different way of computing, a different paradigm.
It's using some of the properties of quantum physics to hopefully speed up some calculations that are currently intractable on the current class of computers that we have. Some people used to call that like the second quantum wave revolution, because already with the current computers that we have, we already use quantum mechanics, transistors.
But now we want to use, to leverage other properties of quantum system, either entanglement, superposition, quantum tunneling, to yeah, have some speed up on some algorithm like Shor's algorithm, for example. So it's a totally different kind of paradigm. Now a artificial intelligence is in general, basically thinking about having an intelligence that is not human that is able to perform human-like tasks. Right, and under it, you can actually write some algorithms that we do that.
You have machine learning and you can have neural networks, that kind of generally people think of it like representation of the brain, even though sometimes it's not like that. Even though it's remarkable to see that some of the intuition behind things like conversational knowledge works is basically how we see, to basically come with the design of that kind of deep neural networks, basically, to be able to do image recognition, for example.
So that's basically two different communities and a lot of sub-fields within those communities. And now within the sub-fields, yeah, you can find some correlations. I will tell you, for example, one of the correlation that I'm mostly familiar with in simulating quantum many-body systems on what we call classical architectures like your laptop or whatever cluster we are using here at Perimeter, right? So for us to be able to simulate quantum many-body systems are different methods.
One of the popular methods is a quantum Monte Carlo method called variational Monte Carlo. To use that, you need to be able to have what is called an ansatz, which is just a good guess of what the ground state wave function of your quantum many-body system is. But to have this good guess, you need to understand the Hamiltonian or the physics of the problem at hand. Is it fermions? Is it bosons, right? What are the interaction strength? What is the Hilbert space? Is it a Fock space, right?
And based on that, on the symmetries of the system, you come up with a good ansatz. Now, not everybody can do that, right? We really need very specialized knowledge. And the moment you perturb the Hamiltonian, that you go to another Hamiltonian, maybe it's totally out of your field. If you leave fermions and go to bosons, you don't get the intuition anymore. So the idea of neural networks, that is borrowing like some knowledge from neural networks.
Since there are universal approximators, and hopefully they should be able to represent any kind of function, then why not representing then the ground state wave function of a many-body system. That was the original idea of borrowing this kind of neural network, basically perform quantum many-body simulations.
And even though nowadays we see that we still need a little bit of quantum intuition to make it work perfectly, like you need knowledge of symmetries, for example, we encode it in a neural network to make it represent your system in a much better way. But yeah, so the story is that, yeah, we saw how it was working amazingly well in machine learning. And it is also starting to work quite well. - You and I, Estelle, we actually work in similar research areas.
You kind of said already, we're both part of this Perimeter Institute Quantum Intelligence Lab. We have our matching hoodies today. - Green hoodies. That's because the acronym for that institute is - Is PIQuILs. - PIQuILs. - They have to be, everything's green at the PIQuIL. - PIQuIL.
- Yes. (all laughing) - And I think something that's pretty unique about this group, at least compared to maybe other research groups at Perimeter is that there tends to be a lot of opportunities for collaborations with industry. So can you talk a little bit about that? And what maybe could be unique or what's important about these academic and industry collaboration? - Definitively. What is unique first of all is the field.
The field, as I mentioned, we are using a lot of state of machine learning techniques, which we know industry use a lot. Facebook, Google, they have huge research groups that publish a lot of papers. So already in that sense, we, just by using those tools, we are already somehow in between industry research and academia research. - Are those classical machine learning techniques? - Those are classical machine learnings.
Even though now a lot of those big companies are having quantum groups as well, and they are developing quantum machine learning techniques as well, and a lot of startups as well. So the field of quantum computing is being pushed forward, both by academia and industry. And the PIQuIL is trying to bridge, I mean, those two worlds and to provide a platform where academia can talk to industry and vice versa, and together working on the projects we can speed with which we advance things.
- I often think of it like an area with many bridges, right? Because you're trying to bridge academia and industry, but also quantum with machine learning. - Exactly. - Lots of different bridges you have to go over. - Exactly, exactly. And one interesting thing that has come up in the last few years is physicists are thinking of actually importing some of the methods that we have been using to quantum matter to the machine learning community. I think of tensor networkss, for example.
They're like, oh, we have a very good understanding on these tensor networks. We can interpret them instead of using your black boxes. So maybe you could use that for, I don't know, image recognition. And people have been doing that and it's working. So it is also a way for the physics community to somehow give back to the AI community. - You mentioned that what brought you to Perimeter in the first place was looking at Roger Melko's work.
And now Roger is, he's the head of PIQuIL, the quantum intelligence lab. Can you just give us a sense of what it's like at PIQuIL? What is a day like at the PIQuIL? What are the sort of questions and problems that are being tackled there? - PIQuIL is really like a startup like kind of environment. Even though there's industry and academia there, there's a lot of free discussions. We have journal clubs. It was virtual during COVID. Now we are starting to come back person.
A lot of discussions about Slack, "Oh, this is a new paper. What do you think of? Oh, I have a problem in my research. Do you have a solution for that" and things like that. So really a lot of interaction. - And so you first came here to Perimeter maybe to pursue more the academic side of things, but as time has gone on, you've become more and more involved with industry. And now you're actually the Co-Founder and the Chief Technology Officer of a company called yiyaniQ.
Can you tell us a little bit about your company and what it's trying to do? - Definitely, maybe I will take a step back by a little bit. I was doing my PhD and then I was doing my post doc. So I was mostly focused on academic work. But even though I was focused on that, my specialty is developing algorithms to solve optimization beside probing the behavior of quantum many-body system, but optimization points that are like real-world problems. But typically the way we solve it is okay.
Like physicists, we like to have like a very easy model that we can benchmark and things like that. That is not really reality. It is not gonna affect the life of somebody. And so I always had behind my mind, in the back of my mind that these algorithms, we could actually try to use them to solve real-world problems, not just write it at the end of the conclusions of our papers. And, oh, you can use it to solve a real-world. So I had that in back of my mind.
And yeah, so last, I think one year and a half, we had these very nice results of an algorithm we designed. And we decided to basically file a patent away. And that was the moment I was like, okay, now we need to try to commercialize it and see whether we can have real-world impact. And we created yiyaniQ.
So the company right now is focusing on designing what we call quantum intelligent algorithm to basically speed up derivative pricing, which is a specific problem in quantitative finance, in the sales side of financial market. In the beginning, I was very much confused. I had a hammer, I didn't know where I'd find the nail. So there are so many different optimization point out there. Some are very interesting. Some are very challenging. Others are boring.
I really needed to find one that was challenging enough, but I found that very fast that, yeah, you need somebody who has expertise to be able to design that. And I met him in an incubator called Creative Disruption Lab, Behnam Javanparast. And he has a PhD in theoretical physics, in condensed matter as well. So we could talk to each other, but he also worked in a bank for more than seven years. So it was quite very easy for us to kind bring our perspectives to found yiyaniQ.
- I'm hoping you can tell us a bit more about optimization problems generally. Could you tell what the term means and how you apply your techniques to it? - Usually for us physicists, it is useful for us to kind of map a problem into a configuration that we understand best. And one sweet thing is that we can view optimization problems as a search problem in a very complex landscape, where in an optimization problem, typically you have a function you want to minimize.
Everybody more or less understands functions, but for a physicist, I can see that function as an Hamiltonian. Directly when you tell me Hamiltonian, I was like, yay. I have a lot of tools in my toolbox to be able to deal with that. And I can view the Hamiltonian as a landscape. You could imagine, for example, in the Himalayas, you have a lot of hills and valleys, can be kind of very crazy landscape.
And solving the optimization problem means from a physics standpoint is finding the ground state of the Hamiltonian that represents that optimization problem. But from a graphical point of view, it means finding the deepest valley in that mountain. And for you to find the deepest valley, you need to search, go up and down. And depending on how you search, you can be more efficient in finding the landscape.
But if your landscape, for example, has a lot of valleys, a lot of saddle points, it has tall hills, right, and maybe very wide hills, it might be difficult for you to be able to find the deepest valleys. This is hardest search problem where solving on optimization problem would be seen.
- Would it be similar to if you wanted to find the deepest valley in the Himalayas, you could walk up and down all of these things, but optimization is a way, is an attempt to not put in that sort of brute force work, but find the simplest route to the answer. - Exactly, it's finding the simplest route to the answer, which definitely what you just described going up and down could be mimicked with algorithms. And it has been mimicked with algorithm.
The most notable one is simulated annealing, where going up and down is having some thermal energy to basically overcome barrier till hopefully, basically you find the deepest minimum. But imagine that you're going up and down with your car. Some moment, I mean, fuel is gone. What do you do? So in the simulation is when you are ramping down the temperature, and then yeah, there's no temperature, no fuel, which means no fuel, no kinetic energy. And then you get stuck in a local minimum, right?
But you could think of a different paradigm which people have thought of using quantum computers or using one property of quantum system that is called quantum tunneling. Right? Then instead of going up and down the valley, you basically tunnel through the hills in the search of the deepest minimum. And then that hopefully will be a faster way for you to find the deepest minimum.
This is not a crazy intuition, because when you think about the way we build tunnels nowadays, if you're like a building company and they say, okay, you need to build like either rail tracks or you need to build a road through the mountain, if you see that the mountain is for example, very tall, but then the width is not that long, you're not gonna build these tracks on top of the mountain. That doesn't make sense. You build a tunnel, quantum tunneling. So that's kind of the idea.
But at the same time, if your mountain, the height is not that high but it has a very like long width, doesn't make sense to build a tunnel. You just go over it. So classically it's better. So that's the reason why most of the time, people do not care whether quantum tunneling or quantum annealing or classical learning is better. It totally depends on the shape of the landscape, and the shape of the landscape depends on the hardness of the problem.
- You told us that your company, yiyaniQ, its main focus is using these techniques on the problem of pricing derivatives. And that's a financial markets term that I barely understand. I believe derivatives are contracts between financial institutions that are based on assets within these contracts. That's about all I know, but it's a difficult problem. Pricing derivatives, I know, is a very difficult thing.
I'm hoping you can tell us why it's difficult, how it's currently done and how you hope to do it better and more efficiently. - Yeah, that's a very, very good question. Indeed, like we are focusing on what is called over-the-counter derivatives that are mainly traded by very big financial institutions. And some of them, they're called like structured products, they are quite complicated to price. So the way it's currently being done is using Markov chain Monte Carlo.
And for you to be able to price them, you need to come up with a large number of possible financial scenarios that obeys the law of large numbers. So the variance of your estimator, of your price, goes down with one over the square root of the number of scenarios that you can generate. So basically you need to generate a lot of scenarios to come up with an accuracy that satisfies a trader, for example. That takes a lot of time. So we talked to some traders working at banks.
They told us that some of the books that have a lot of underlying products in one contract can take from 60 to 90 minutes time to price, and they need to price it a lot of times during the day, every day. So not only it takes a lot of time, since they have, like they cannot go beyond a certain amount of time, which means they cannot price a certain number of scenarios, they have to reduce the number of scenarios of price. It means they cannot have the margin that they expect.
So they told us that sometimes they could be mishedge of $10 million. That's the error bar of price. - $10 million error bar. - Exactly. - I wish I had that error bar. - That's very huge. - Well, it depends which direction it's in. (all laughing) - That's very huge. So idea is basically because we know there are some methods that are more efficient than Markov chain Monte Carlo, be able to price it faster and also more accurate. This is what we are hoping to achieve.
So typically you want to find the deepest valley, but sometimes it's very hard. So if you want, you find a valley that is not so far from the deepest valley, you're fine. That's like they call it near optimal solutions. That would be fine as well. Say for example, you're solving the traveling salesman premise, as you mentioned, if you don't find optimal path, okay, the salesman will not be angry if you find a near optimal path. That saves him time and money. - Right. - Right?
- He probably won't know that it's not the actual optimal path. - (laughs) He probably won't know. - Yeah, that problem essentially is how does a traveling salesperson hit a certain number of cities in the most efficient way possible. And it's just a very difficult mathematical problem, right, an optimization problem. - Yeah, definitely, definitely, definitely. So if it's not exactly solved, but approximately solved.
So for the financial case, what we are trying to do, so the crucial part of our approach is that we need to be able to model the financial problem of derivative pricing as an optimization. And then we can use quantum annealers. We can use all kind of flavor of simulated annealing, parallel tempering, whatever it is. We can use variational annealing. We can use mem, just variational optimization with neural network.
So that's where you really need the financial expertise to be able to cast it as an optimization problem. That's our approach, which is different from the approach people have been having before because we know for example, there are algorithms on the quantum computer to solve the price derivative, like quantum amplitude estimation on measurement-based quantum computers.
But we could use measurement-based quantum computer as well, because we know that the techniques like QA, quantum approximate optimization algorithm can be used to solve an optimization from a CP-based or measurement-based quantum computer. But by looking at the current state of quantum device with the qubit bonds, with the noise level, we feel like for relevant real-world problems, we are not there.
So our approach was mainly focused on an annealing-based approach, plus physics inspired, plus machine learning techniques. - And the name of your company is really interesting. And I'm wondering if you'll share with us the story of what the name means and how you came up with it. - Definitely, so I like the PIQuIL so much and the fact that our kind of motto is kind of quantum intelligence.
So I wanted to have something similar, but in my local language, to be innovative and to differentiate self from everybody else. But I don't speak my local language very well (laughs). So I kind of, I asked my whole family, my mom, my dad, my brothers and my uncle and aunts to come up with a name that means quantum intelligence in my local tongue called Basaa. First they told me that quantum, they don't know what it is, (Lauren laughing) even in English (laughs).
So we kind of put it out of the picture. I told them, okay, something like shell intelligence, intelligence of the future, something like that. They came up with different names and my mom won. She came with yiyani. Yi, that means intelligence, and yani tomorrow, which means the intelligence of the future basically, and the Q at the end. - So intelligence, future, quantum, it seems like a pretty great name for what you're doing. - Yeah, yeah, definitely. - Yeah.
- Do you remember any of the names that didn't make the cut? - No, my God, so many. (all laughing) - And so I know your company has grown a lot, as you alluded to through this Creative Destruction Lab program. Could you tell us a little bit more about this program? - Yes, so basically it is like an incubator for quantum companies. In fact, they had a number that about 25% of the quantum computing companies passed through their program, can you imagine, in the whole world.
So it's really like one of the main incubators of quantum computing companies. I knew about it before, because Roger is very much involved. I think he's the academic director of CDL. So I already knew about that. And when I decided to create a company, I applied for the boot camp. So they have a boot camp usually during the summer for about a month half-ish. And so I went there. There are a lot of course fundamentals of quantum computing, quantum physics.
What are the current states of quantum architectures? There are so many different way of building a qubit. What are the current business cases? What are the potential advantages and things like that. And then you have a world core of quantum enthusiasts. You could start a company or you could, because some of them are startup, you could join a company. I got a lot like offers for example, during the bootcamp.
But then yeah, so the idea of that is basically helping people who have ideas on using quantum computing technology to solve real-world problems, to basically groom them, help them navigating the landscape. - And I know you have a lot of experience working in the academic side. But probably working in industry, I guess there's a whole new skill set that comes with working in this new field.
Were there any lessons that were particularly useful from this camp, as you tried to build this bridge between academia and industry? - Definitively, I still want to do research. For me, the most shocking truth is that businesses don't think like researchers. I learned that they don't care whether you're using state of art technology or new technology. They just want you to solve a problem.
And so for me, when I think about, oh, if I, for example, improve an algorithm of an order or two order of magnitudes, I'm excited. If it does not translate into them earning more money, they don't care about that, (chuckles) right? So it makes me have a different approach on doing research for business. I have to do research, yes. I need to think about using the best possible tools, yes. But at the same time, I need to think about potential business advantage, which we don't think about.
Of course, we don't think about that. We are most interested in solving exciting problems. - It's like optimizing a different function. - Exactly. (Estelle and Lauren laughing) - Was the term boot camp applicable? Was it pretty intense? - Oh yeah, it was like, it was crazy. And in fact, the craziest time of the boot camp was it had a two day hackathon. I think I probably slept like three hours during those two days.
You had to come up with an idea to solve a relevant business problem using a quantum. - In two days. - In two days. - And any problem, or they told you a certain problem? - Any problem of your choice. So they had some problems, that maybe some hints, but any problem using some of the architectures that were made available to us. And yeah, program it and come up with results. So there was only, not only the scientific value.
You need to come up with a business pitch, like do some quick market research, show that, come up with the numbers that this is a relevant problem and have a short video of making your pitch. - Hang on, you've got two days to develop quantum algorithms and a business pitch and a video. - Yes. - Okay. So when did you get those three hours of sleep? (all laughing) - I was working with Behnam until midnight, I think. (Colin and Estelle laughing) - Did you just crash at the end?
- It's when we stopped talking around maybe midnight or 1:00, and then I kept working till probably 3:00 and got up at 6:00, and started working again. - And what did you actually end up developing? - Oh, we basically wrote a code on the D-Wave machine to solve a portfolio optimization. And we had to push it on GitHub. So it's available on CDL GitHub.
- So hold on, not only did you have to come up with an algorithm and a business plan, but then you had to push this out and make it available to other people. - Yeah, publicly available, yeah. - And you mentioned D-Wave. Can you explain a little bit about what that is? - Oh yes, so D-Wave is a quantum computing company. It was the first one to actually commercialize the quantum computer. And so they are mostly focused on annealing-based approach as is solving optimization.
Even though recently, they announced that they are starting to build also CP-based quantum computers. So one of the cool thing that they did, and a lot of quantum computing companies are doing now is if I want to run simulations on a quantum computer, I don't need to go and buy 10 million, whatever the cost is, - Thank goodness. - And come and install it at Perimeter. You can have access to it through cloud.
And so you have an API code and you just, yeah, pass in parameters and it spits you back basically the results. And you can even see which qubit you have been using the quantum processor to basically solve your problem. - So you can implement your algorithms on D-Wave, but in the cloud you can do it from anywhere? - Definitely, yeah. - It's amazing. - Oh, not anywhere. It depends on where they have the clouds deployed. I think now you can do it in North America and Europe.
South America, I'm not so sure. Africa, I'm not so sure. Probably in Japan as well. So as they are expanding, they provide that cloud service. - And as you've said, Estelle, it seems like there's just so many different priorities that you have to balance when you're doing this work at kind of the intersection of academia and industry. And we had a grad student from here in Waterloo send in a question about that. - This is Matthew Duschenes, a student at IQC and Perimeter.
I'm wondering, how do you balance coming up with novel research ideas versus staying focused on your specific startup objectives? - Nice question, very, very important question. I ask myself that question every single day. (Estelle and Lauren laughing) - Are you able to balance these things or is it always a juggling act? - In the beginning, it was so hard. It was really, really, really very hard.
Now I'm kind of equilibrating roles, dividing my time half, half, not every week, but yeah, that's what I'm trying to do. Because for the company, definitively I'm doing like an application of my techniques, but we are in a very fast-paced milieu, whereby you need to be aware of whatever is state of art. So you need to be on top of your game as far as research is concerned. So I need to keep an open eye on the research world as well. That's why. - Must be changing every day.
- Exactly. So it's not as before that I could read archive paper every morning. I cannot do that anymore. (Estelle and Colin laughing). I can attend journal clubs, attend conferences, and I talk to collaborators to keep in touch with what is happening as far as research is concerned. I was groomed as a PhD student that a problem is interesting when it's hard. I mean, if it is not hard, what's the point (laughs)?
So I really like taking on very hard problems and if they're relevant to an everyday person. - When you get stuck on a really hard problem, what do you do to push through it, to get past that obstacle? - I do it very badly, usually (laughs), almost depressed. Anyhow, but yeah, typically just take a step back and try to do something else. I mean, go boxing. - Boxing? - Go swimming or running, something different.
Sometimes involves talking to collaborators to get some of the ideas that they have and coming back to it with fresh eyes. - So you've been telling us a lot of stories of things that have happened in the last few years. And I'm wondering if we can maybe go back a little bit further. Could you tell us the story of how you first got into being a scientist or how you first decided to pursue that type of path? - I have a very non-typical path to becoming a scientist.
Yeah, so right away I should make a disclaimer. I didn't plan to be a physicist (laughs). - This wasn't the lifelong dream. - Nope, it wasn't a lifelong one. Yeah. Well, I wanted to do naval architecture. So I was advised during my high school that for me to do naval architecture, I needed to have a bachelor in physics. So I got. - Sorry, naval architecture is designing ships? - Ships. - Okay. - Yeah. It's very, very different.
- But maybe you're gonna use your methods for naval architecture next? I guess we'll see. - Yeah, why not? - Maybe there's an optimization problem in naval architecture. - Oh, naval architecture, I don't know, but definitely in the maritime industry on ship route, there is an opportunity for that. I even thought about that, either ship route or ship loading. For example, imagine I have a big cargo. It has to load on thousands of different containers on the cargo.
What is the best way for you to do that to optimize the space? I actually wrote an algorithm that, VNA, yeah. - Shipbuilding and ship architecture, where did this come from? - Since my mom worked in the maritime industry, I was very much influenced by her. So I wanted to do a job that was related to sea, ocean, right? But I wanted to do a technical job, something that I could use some of the things I was interested in, mathematics, physics.
So I found that naval architecture was the best, but I was not well-advised. I found out that you cannot do naval architecture with a bachelor in physics. Then I wanted to do computer science after when I found out I couldn't be a naval architect. But unfortunately that year in my homeschool, they didn't open up a master in computer science. The only available master was in physics. So I was like, okay, I need to go to school. Let me just really start for the master in physics. And I like it.
It was very easy for me to do, and I think I got first class and then I got a scholarship to go to Italy to do it. - It sounds like you're still interested in shipping and ships. Is that an ongoing fascination for you, the maritime industry? - No, I think it after I was so disappointed, I should say. I was really, really, really disappointed when I found out I was just missing that, so that kind of died out. But for computer science, yeah, I'm mostly programming now.
Almost all of my day I'm writing code. I kind of brought together my interest in computer science and programming in my physics job. - Well, we actually got a question about how to combine programming with research in physics. So could we play the next question? - Hi, so I'm Hassan Conser from India, and my question is a little more career-related.
How do the fields of programming and physics mix like simulation machine learning, and is it necessary to learn programming when going into field of physics? - From the first part of your question, if you think about machine learning for physics, you definitely need programming for that, right? But if you have to think about physics in general, it really depends on which field of physics.
There are some fields of physics where not a heavy amount of programming is needed, some even none, just need to do some kind of analytical work. But when you think about the field of physics, generally as a rule, my feeling is that little bit skill on just knowing how to plot functions is important. Just knowing Python, which is very easy to learn, should be sufficient to get by. But if you want in field of computational physics, yes, you need to know how to program a little bit more.
And nowadays it's really easy. For example, for machine learning, there are a lot of libraries you can just use, I think about 10 or so, to write prototype of your model and to test it very quickly. You have things like Google Colab. You can use GPUs to simulate very fast things and even get some results. So I feel like it shouldn't be seen as a very huge barrier. Programming is actually very fun.
But my advice is that you shouldn't lose sight of the fact that at the end of the day you're a physicist. So you need to sharpen your physical intuition. I give you the advice one of my lecturer gave me when I was doing my PhDs. You first of all need to take your pen and paper and figure out the physics behind the problem. And once you do that, then yeah, you can take your computer and write some code. - You've used the term physical intuition a couple of times.
I'm hoping you can explain what you mean by that. - Physical intuition is based, I would say, on the understanding on how nature works and the understanding of some physical principle. Take like the Heisenberg principle for quantum mechanics. If you know exactly the position of a particle, you cannot know exactly momentum of that particle. So when you think about a problem, you need to have these kind of things on the back of your mind, and that will help you not only interpret results.
It will help you design models to maybe benchmark something specific about the model. It's very, very important to do top class research. That's my impression. - You've attended workshops in Africa about promoting women in science and just promoting science in Africa overall. Can you tell us why you wanted to attend those and what you hope people got out of your presentation or your attendance? - I always find myself living in a superposition of two almost orthogonal worlds.
Unfortunately we know that science in Africa is a little bit lagging compared to the West, but for women it's even worse. It's like this really was because a lot of cultural apprehension. It's changing, it's really changing, but still sending women to do what is called hard skills, typically people think that, okay, math, physics, it's just for men, even when they're trying to be progressive. Even here in the West for women, we see that as you go up the ladder, we see less and less women.
It's even stronger in Africa because there's more commitment that is demanded. And the role of the woman in family, it demands a lot of your time. That makes it very hard for you to do top class research. Starting to have these conversations, one of the feeling that I've been having, that it has to start first with women scientists, African women scientists, the mindset, to kind of recalibrate the mindset that it is possible for me to do science. I don't necessarily need to create.
It is possible, and from there, like put together policies. I feel like this is very important. Also educate our male counterparts, starting with our families to really change that mindset. But me in my career, I had a lot of instances of people telling me that, why are you doing your PhD? You should be married and having kids and preparing for your husband, (laughs) right, this kind of a thing.
But I was educated in the house when my mom told me that as a female, you can do whatever a man can do. I already had that in my mindset, but other people, they don't hear that kind of thought. It can really affect them. But starting to have this conversation, we hope to see change. - Is that what you're trying to do when you go there? You're trying to help with this recalibration process for individuals. - Even not necessarily when I go there.
Whenever I happen to interact with female scientists from Africa, which happened once in a while, yes, trying to have those conversations, change of the mindset. It doesn't have to be this way. - I read that your father, he'd wanted to get a PhD in physics, but he didn't 'cause there are more practical paths. He chose engineering, I believe? - Yes, actually it was not too much of his willing. So my father was, he is still very smart. He was very smart.
So he had a government grant after his high school degree to go to France, to basically to study. He was studying physics. And then he wanted to do a PhD in physics, but the government was paying his stipend. It's like, we don't need physicists. We need engineers. For him not to lose his scholarship, he had to move to engineering. But then he really encouraged me a lot to do physics. - So what was his reaction when you obtained your PhD in physics? - He was very happy.
In fact, he told a story during my PhD party of the fact that when I was doing first year bachelor in physics back home. So I did in high school, French education. I studied in French. And naturally I'm like French speaker, my mother tongue, willing God. But then I moved in English in the Western part of Cameroon. It was very, very hard. I needed first to understand the English before understanding the physics. It was, I had a dictionary all the time when I was going to the lecture.
So it was really, really bad. One month after starting my bachelor in physics, I passed an engineering concours in the French side of Cameroon to become an engineer. So I called my dad. I was like "I'm stopping this thing. It's not going; I need to go and do engineering." My dad told me that "No, Francophones have been able to go to that school and graduate. You're gonna stay there." I was so mad at my dad. I was so angry.
But after a couple of months, I picked up the English and I did very well. And he told the story during my PhD party that I hold strong and now she's a PhD; she's a doctor. So that was sweet. - That's beautiful. - And Estelle, you've told us so many nice pieces of your story starting in childhood.
And I know you kind of have alluded to the fact of how you made the decision to come here to Perimeter for a postdoc after your PhD. But I know that you actually had a lot of options for what to do after you had a PhD. And I always look back fondly when you were making that decision 'cause you and I actually talked before you came here. And so I always like to tell people I was one of the first to meet you here at Perimeter.
So could you tell us a little bit more about how you made that decision? - Yes, definitely. So one thing I wanted to make sure is that people I would be working with, especially Roger Melko, I already knew he is a great scientist, but I wanted to know that he's a good person to work with. So I wrote to you and sent you an email and you were very nice to have a Skype discussion with me. And I was just convinced. There's also Giacomo who sent me an email, who replied to my email.
He told me that it's amazing to work with Roger. That convinced me that PI is a great place, but at the same time I had opportunities, one in Alberta, but they called, canceled it out. There was one at Microsoft, which was actually the most interesting one.
I had one also in California, which was kind of interesting because we have collaboration with people at NASA, but then I'd be working mostly on developing further the algorithms that I learned during my PhD. At Microsoft, I would've been applying the algorithm I developed during my PhD on WeWork programs. That was extremely exciting for me, but I wouldn't have learned something really new, would have been mostly application.
Whereas here at Perimeter, I would have enlarged my research interest to include machine learning and neural networks. So that is basically the reason why I chose to come to PI. And I don't regret that at all. - And Estelle, now at Perimeter, your title is Research Scientist, but when you first came here, you had the title of a post-doctoral fellow under the name Francis Kofi Allotey Fellowship. Can you tell us a little bit about this fellowship and how it was named?
- So Francis Kofi Allotey, unfortunately he passed away about five years ago. He was really a monument of an African scientist who literally inspired and trained generations of physicists on the African continent. So he actually did a graduate degree at the Imperial College London under the Nobel Prize Winner, Abdus Salam, who later on created this famous, the Abdus Salam International Center for Theoretical Physics. And then he did his PhD at Princeton under Robert Oppenheimer.
- He did his PhD with Robert Oppenheimer. - Yeah, so he was the first Ghanaian to do almost everything, the first Ghanaian to earn a PhD in mathematical physics, the first full professor in mathematical physics in Ghana. And as far as research is concerned, he is mostly known for this Allotey formalism, which is basically a way to detect soft x-rays in material like lithium or other alkaline materials. And yeah, so he kind of, I mean, he has a single authored paper on that, which is pretty neat.
And he got, I think, a medal for that. But beside his research contribution he had, he was member of a lot of international bodies. He created, was one of the founding member of the African Physical Society. He was a board member at ICTP and a lot of other institutions. And he did a lot of work as well in Ghana, like creating institutes and fostering like science in the continent.
So it was quite an honor for me to have a fellowship named after him, almost more than my shoulders could bear, right? But it was good also for people in the West to see that, because typically you're not familiar with that kind of a name. You think more like Einstein, Dirac, Schrodinger, but it's good to see that we also have, if you want, we can groom up top class scientists. So to how did I come up with the name?
So when I was coming to PI, Neil Turok actually gave me the choice, interestingly enough, to choose which name I wanted. And yeah, I chose him. - Now, you've kind of expanded this set of tools that you have through your postdoc. And now you're getting to work on maybe some of these real-world applications.
And I really liked the thing you said earlier about how often in papers academics are claiming that their methods could be applicable to all of these potential, huge, real-world problems, but maybe people don't always really put the effort into solving those. And it seems like that's really what you're trying to do now in your work, which is really impressive. And I'm just curious what other real problems you have in mind to look at next? - Oh yeah, maybe not to look at next.
Kind of the biggest problem I have in mind to solve is protein folding. - Protein folding? - Yes. - What is protein folding (chuckles)? - So basically a protein to be functioning, it has to have a certain conformation. It can take millions different conformation. When you're doing protein design, you need to find a configuration white box. They call it like when the protein is native, in its native state or in its folded state. And usually it starts from an unfolded state.
And the path through the folded state is like you going through the Himalayas, is a very, very hard path. - Peaks and valleys? - Exactly, exactly, with a lot of like local minimal saddle points. It's very hard, and not only is it a hard problem, it's very relevant, drug design to help us like have better drugs and help in the health sector. So this is also, it's having a very strong impact, but it's also very hard to solve. So this is one of the problems that I'm thinking.
I know that the approach we are having now is really to develop state of optimization algorithms, that of course, in the company right now we apply it in the financial domain, but that we can easily export to other domains, like in the domain of protein folding where the only subroutine or function we will have to change is just the Hamiltonian. But then we will need to have domain knowledge. And how do you write the protein folding Hamiltonian?
And that actually is not a minimalistic model, is not an easy model, but it's really a model of a real-world model. I cannot; you really need somebody who works in either quantum chemistry or biotechnological sector and things to be. - Just to see if I'm understanding, right now when you're working, Behnam, your co-founder is kind of bringing this expertise of the finance side. So you would kind of need someone analogous to Behnam for this protein folding problem. - Exactly, definitely.
So we even talked about that with Behnam, that once we will be able to create value with those algorithms, we start exploring what we call in business jargon other verticals. - Well, it seems like there's a lot of potential options for the future, and I'm really excited to see what you're gonna optimize next. - Thanks. - Thanks for chatting with us today. It was really fun. - My pleasure. (upbeat music) - Thanks so much for stepping inside the Perimeter.
Be sure to subscribe so you don't miss a conversation. We've interviewed a lot of really brilliant scientists whose research spans from the quantum to the cosmos. And we can't wait for you to hear more. And if you like what you hear, please give us a rating or a review wherever you get your podcasts. Great science is for everyone, so help us spread the word.