Jon Krohn: 00:00
This is episode number 495 with Greg Coquillo technology manager at Amazon and 2020 LinkedIn top voice for AI and data science.
Jon Krohn: 00:12
Welcome to the SuperDataScience Podcast. My name is Jon Krohn, chief data scientist and best-selling author on deep learning. Each week, we bring you inspiring people and ideas to help you build a successful career in data science. Thanks for being here today, and now let’s make the complex simple.
Jon Krohn: 00:42
Welcome back to the SuperDataScience Podcast. I’m delighted to be joined today by the concise and articulate Gary-Gregoire Coquillo, but you may know him simply as Greg Coquillo, LinkedIn’s raining top voice for AI and data science. When he’s not sharing succinct summaries of both technically oriented and commercially oriented AI developments with his 70,000 LinkedIn followers, Greg’s a technology manager at Amazon’s global headquarters in Seattle. Originally from Haiti, Greg obtained his degrees in industrial engineering and engineering management from the University of Florida before settling into a series of management-level process engineering roles. Today’s episode is well-suited to listeners at any stage in their data science journey. As even while discussing cutting-edge approaches like quantum machine learning we stay relatively high level. If you manage commercial AI projects or aspire to today’s episode will be of particular interest since a recurring focus in the episode is on how to get a return on investment in AI projects and in AI startups. All right, you ready for another awesome episode? Let’s go.
Jon Krohn: 01:58
Greg, welcome to the program. I’m so excited to have you here. I’ve heard so much about you and I can’t wait to learn from you on the show. How are you doing and where are you calling in from?
Greg Coquillo: 02:10
Jon, thank you so much for having me. I’m super happy to be here as well. I’m calling from Seattle Washington. So yeah, really nice.
Jon Krohn: 02:21
How’s it going out there? Pandemic starting to go away, things are opening up?
Greg Coquillo: 02:27
Yeah, absolutely. We’re starting to see an adjustment in people’s normal lives. So we’re starting to see the normal things coming out. So restaurants are opening.
Jon Krohn: 02:38
Nice.
Greg Coquillo: 02:39
The gyms are opening, no problem. Simply a lot of people are taking the chances of getting vaccinated, doing the right thing off course. So it’s great to see, to feel like you don’t have to get so rigid with the rules. So I’m feeling a little bit more normal now. That side is more traffic now, so.
Jon Krohn: 02:59
Yeah, exactly. There’s definitely been a few great things about the pandemic. Here in New York, especially in the beginning, April 2020, I could run around, bike around anywhere I wanted. There was never any traffic. If you needed to get somewhere in a car, [crosstalk 00:03:15].
Greg Coquillo: 03:17
Exactly.
Jon Krohn: 03:17
And now we’re back to our six mile an hour slog.
Greg Coquillo: 03:22
Yeah. It takes me 25 minutes to get to the office in which traffic you can triple that time easy.
Jon Krohn: 03:32
I’ve only been out to Seattle once, but I really enjoyed it. I was there for one night. I’d been skiing and, man, I can’t even remember where I went skiing, but there’s was a huge snow storm, couple of years ago, maybe a year before the pandemic. And wow, I can’t believe I can’t remember the mountain, but there’s beautiful skiing out there in the Rockies. And it was about a two hour drive into Seattle and I flew out of there. So stayed one night. I don’t know. Didn’t really see much of the city.
Greg Coquillo: 04:00
Oh, yeah.
Jon Krohn: 04:00
But it seems like a really nice place to live.
Greg Coquillo: 04:03
[inaudible 00:04:03] because when I showed up here, it was right in the middle of pandemic. Two weeks later it was like, hey, everybody needs to stay home. Can’t see anyone. So really my training at work was remote and I never had a chance to explore. So hopefully by the end of this year and into next year we’ll get to explore a little bit more. So that’s one thing I’m looking forward to.
Jon Krohn: 04:29
Nice. So you moved out to Seattle to work at Amazon?
Greg Coquillo: 04:33
Yes.
Jon Krohn: 04:34
And that’s their biggest office, right?
Greg Coquillo: 04:39
Yeah, main office in Seattle.
Jon Krohn: 04:41
And so we’ll talk a little bit about your work, but first I want to talk about how we know each other. So Harpreet Sahota who was on episode 457. I kind of was aware of you through him because you show up in his happy hours. I haven’t been to his happy hours, but I keep hosts photos from there, and so I see your face. And so you’ve been on my radar for a little while. I was like, I can’t wait to like have the opportunity to talk to Greg and see if he wants to be on the SuperDataScience Podcast. And then you were at DataScienceGO in April. So DataScienceGO is the conference affiliate of the SuperDataScience Podcast.
Jon Krohn: 05:27
And I reached out to do the DataScienceGO organizers. And I said, “Have you had any speakers recently? They were absolutely amazing, because I’d love to get them on the SuperDataScience show.” They recommended you. And I was like, “Great.” I asked you right away because I was like, I’ve known about Greg for months. And now finally I get to have you on the show. Oh, you’ve also been on Harpreet Podcast, right? You’ve been on his Artists of Data Science Podcast. You’re talking about product management.
Greg Coquillo: 05:56
Yeah, that’s right. It’s such an honor. I thank DataScienceGO team for this referral. I’m truly humbled by that. And who can miss Harpreet happy hour? I can’t, I can’t afford to miss this. Right. So I try my best to be there every Friday. It’s such a very great environment to discuss data science for data science folks and non data science folks. So it’s very welcoming. So I truly appreciate being there and learning from everyone participating there. And thanks for finding me there too. So now I’m on the SuperDataScience.
Jon Krohn: 06:34
Yeah, you’re here, it’s happening.
Greg Coquillo: 06:38
Yeah, that’s great.
Jon Krohn: 06:40
So you are a prolific content creator and this has kind of, see you pop up and then I see you just, I guess it’s like a second degree connection on LinkedIn. Initially I see tons of content because every day you’re posting really valuable content. You post really great data science blog posts. You post academic papers and you write summaries of everything you post. But for academic papers in particular it’s really helpful because you bring home the main point of that paper. You post courses, you post textbooks, and people like it. So you’ve got 70,000 LinkedIn followers at the time of recording by the time this is live, it’s probably going to be way bigger. And you’ve been recognized for it. It isn’t just the quantity of people. It isn’t just me being the host of the show. You were recognized in 2020 as a LinkedIn top voice in data and artificial intelligence. So I don’t know if you’re going to have anything to say, because I’m probably just embarrassing you.
Greg Coquillo: 07:43
Man, I’m starting to feel my head gets so big, but it might explode. But yeah, for me especially research papers for example, right? So they can be a pain to read, but I like the short ones. Short ones, meaning 10 pages or less. And what I do, especially in the summary section of my posts is to provide somebody with much less time to kind of skim through and understand the content of the document, but also give people an idea of what my thought process is after reading the paper. And I’m also hoping myself for somebody providing me an opinion that might be way different than what I’m expressing. And I’m aware that I could be totally wrong. And at the end of the day when I have comments from people after posts, I learn either way. So it’s a great exercise in one, I hope I can take beyond just LinkedIn, because there are other channels also that can harbor these type of content that people can take advantage of.
Jon Krohn: 08:54
Nice. Well, yeah, I greatly appreciate what you’re doing for the community. And of course, like we do for all guests we’ll have links to your profile in the show notes. And at the end of the program, we’ll talk about all the ways that people can stay in touch with you. I actually know with you, there’s more ways than usual, including lots of opportunities to speak directly with you in any given week. So that’s really exciting. People will have to hang around to hear about that.
Greg Coquillo: 09:19
Absolutely.
Jon Krohn: 09:22
So we just talked about how you moved to Seattle in the pandemic. So relatively recently started working at Amazon, but you also had other roles you’ve been working for years, working on kind of process engineering and optimization, right? So at Lonza, at Avery Dennison, and you even have you formally educated in industrial engineering and engineering management. And I actually, I found out just before we started recording that it’s kind of an interesting story as to how you ended up in the US. So you came over to study, but you kind of got stuck here.
Greg Coquillo: 09:59
Stuck in a good way. Yeah, absolutely. And I say it’s some sort of luck that came out of an unfortunate event. And when the earthquake happened in Haiti I was about to end up my master’s degree. And whether a bachelor’s or master’s you have a certain period of time to stay, to look for a job, and have that horror except to sponsor a work visa for you. If you didn’t find anyone that period expires, you would have to go back home. So with an unfortunate event, which was the earthquake, the US government allowed us to stay leveraging a temporary work permit. And that’s what I’ve used to kind of take some stress down, that stress of having that six month period of time to look for a job. Taking my time to look for companies who will accept this temporary work permit to hire me. And that’s how I got into to the professional world, which I’m super grateful for.
Jon Krohn: 11:05
Well, yeah, it’s really great to have you here. Obviously you’re an enormous force in the data science community. I guess that same, at least in terms of the content creation, I guess you could do that from anywhere in the world, but yeah, I think the US is lucky to have you.
Greg Coquillo: 11:27
I appreciate it. Thank you so much.
Jon Krohn: 11:30
So beyond your kind of day-to-day work, I understand that you do lots of homicide. So you provide not only in content creation, not only in your day job at Amazon, but you also can help people out in other ways on the side. So do you want to tell us a bit about that?
Greg Coquillo: 11:52
Yeah, absolutely. I talk to a lot of startups in to help them grow, help them identify use cases, help them identify product market fit, and things like that. So providing some advising to startups is something that I like to do on the side. So I really make my time flexible. It’s amazing when with proper planning inside of 24 hours, how much things you can put out, right? So I try my best to leave a session here and there to help out. And I do have a plan in the near future which is starting to reach out to local universities with these incubators at university level to start offering my help there and start building a network and helping companies, especially AI focused startups to help them kickstart and even channel them through funding as well. So I do have a list of contacts who can help with funding as well. Given that there’s traction, there’s a clear plan for growth. There’s been some historical performance and things like that, depending on what level these startups are. We can find some fundings for them.
Jon Krohn: 13:12
Very cool. So if somebody has an idea, a pitch deck that they put together, they should do just reach out to you.
Greg Coquillo: 13:21
They can, so it will be more of a let’s launch it and let’s get traction before we can start talk about, okay, let’s try to get funding. Because then you need to really walk a little bit because it depends on the people I talk to. Typically, they want to see some traction, some growth and things like that. So before they can bite the bullet, and things like that.
Jon Krohn: 13:42
People definitely want sales.
Greg Coquillo: 13:44
You have to have a team. You have to have the proper metrics. You have to understand the market and things like that and start walking before you start running.
Jon Krohn: 13:57
Nice. This episode is brought to you by SuperDataScience. Yes, our online membership platform for transitioning into data science and the namesake of the podcast itself. In the SuperDataScience platform, we recently launched our new 99 day data scientists study plan. A cheat sheet with a week-by-week instructions to get you started as a data scientist in as few as 15 weeks. Each week, you complete tasks in four categories. The first is SuperDataScience courses to become familiar with the technical foundations of data science. The second is hands-on projects to fill up your portfolio and showcase your knowledge in your job applications. The third is a career toolkit with actions to help you stand out in your job hunting. And the fourth is additional curated resources, such as articles, books, and podcasts, to expand your learning and stay up to date.
Jon Krohn: 14:54
To devise this curriculum we sat down with some of the best data scientists, as well as many of our most successful students, and came up with the ideal 99 day data scientists study plan to teach you everything you need to succeed. So you can skip the planning and simply focus on learning. We believe the program can be completed in 99 days, and we challenge you to do it. Are you ready? Go to SuperDataScience.com/challenge. Download the 99 day study plan and use it with your SuperDataScience subscription to get started as a data scientist in under 100 days. And now let’s get back to this amazing episode.
Jon Krohn: 15:31
Well, that’s really cool. So yeah, another way that you are contributing to everyone to this entire economic ecosystem. It’s brilliant. So not only through startups, but also through your professional work, I gather that you have a lot of experience in making the most of artificial intelligence. So you have a really great sense that you’ve honed over the years of how you can get a return on a commercial investment in an AI project. So do you want to talk a bit about that or maybe kind of your top tips for other people getting a good return on their investment in AI?
Greg Coquillo: 16:14
Yeah, absolutely. So for me AI once you have people… And let me try to make sure I understand the question. And you’re talking about some sort of best framework to get a return on AI investments, right?
Jon Krohn: 16:33
Yeah. If you have something like that, that would be amazing.
Greg Coquillo: 16:35
Yeah. So one thing I keep talking about that really caught my eyes is this company called Element AI.
Jon Krohn: 16:45
Element AI, that’s Yoshua Bengio’s company in Montreal, right?
Greg Coquillo: 16:50
Yes. So they build this AI maturity framework, which I think everybody should use. A lot of times people want to say, oh, I can leverage AI to get myself to the next step. And then they start, and then they get stuck.
Greg Coquillo: 17:05
The first step to me to understand how I can help you is one, realize it’s a tool that can be used amongst different tools. It requires a mindset, a culture, mindset change, and culture that needs to start from the top. Now, another thing too is before you do anything, you have to be able to evaluate your maturity level. So which I’m coming back to Element, which has this great framework. It’s a five by five framework that you can check where you are on the maturity level. So whether you’re exploring, you’re experimenting, you’re formatting, or optimizing or transforming a section of your company or your whole org. And then you have those different section of your organization that you need to measure against that. You have strategy, technology, people, data and governance.
Greg Coquillo: 18:08
So once you understand this, then you need to understand how to separate the use cases. Sometimes we focus on the very short term deliverables, typically that are tied to money which is let’s quickly win more revenues by leveraging AI to better provide recommendations to our users. Or more long-term strategic transformation of your business, which is more like, how can I improve the user experience when they go to my website? How can I improve the customer service experience once we start kickstarting this process and things like that. So those are more long-term transformational things. Once you understand these use cases, you can understand how to go about it with that framework. I think is one of the best that I’ve seen out there.
Jon Krohn: 19:04
Nice. That is a very specific framework that will really help people out. Now, I just realized. And so we were going to get to audience questions later, but I think we’ve actually answered one of the audience people’s questions already. So Bernard Tumanjong whose last name I’m surely mispronouncing. He was a software engineer and machine learning engineer. He reached out to ask you on the podcast what framework should organizations consider to formulate a stable enduring long-term AI vision? And I’m pretty sure you just nailed it.
Greg Coquillo: 19:35
Yeah, I think, I remember seeing something like that. To me it’s the best one. And this is the first step in my opinion to understand what kind of return on investment you want to have. And you have to really learn how to even crawl before you get to project execution and things like that. So, absolutely.
Jon Krohn: 20:00
Nice. All right. So let’s go right into the audience questions. So we had Bernard’s there already. The next one is from Serg Masís, who is someone that I’ve known for years. He was the MC at a talk that I gave at an open data science conference back when conferences were still in person. And he’s the author of Interpretable Machine Learning. He has a great question for you that is relevant to his background, but I think will be relevant to a lot of listeners as well. Which is that Serg is a startup founder. And so he’s interested to know from you, based on your experience advising AI startups, what data science skills or attitude is most needed to succeed with an AI startup?
Greg Coquillo: 20:47
Yeah, I like that question. I think the attitude, especially even skills, we can put all of the professional skills in the box and still apply those. They’re very helpful. We all know them, the kind of skills that a good data science needs. But when it comes to start up, I think in terms of attitude, you have to be able to understand what it takes to move slow and move fast. Right? So when it comes to a startup, just like any other company, you have to have a vision, a mission, a strategy that it explains how you will go about that vision and mission, and also your tactics.
Greg Coquillo: 21:36
So as a data scientist, you have to be good at all of those. Understanding the long-term strategy and also understanding what kind of tactics are available to you, or you can use to reach this longterm or to execute this long-term strategy. And being quick in the tactics helps you iterate fast, help you try different things. And which is the core of a data scientist, right? They try, they’re explorers, they find things, they discover, they try, they confirm things. So being quick on these tactics is what will get you above water as a startup up to future success.
Jon Krohn: 22:18
Nice. That’s a great answer. And I guess with that iterating tactically that’s to solve both your machine learning problems, as well as your business problems, right?
Greg Coquillo: 22:27
Correct. It doesn’t change. It is same thing, as we were saying an AI startup or known AI startup, it’s the same thing. From a business perspective, you want to be quickly iterating because you want to be faster at that point.
Jon Krohn: 22:42
Nice. All right. Well, that was a really clear answer. Thank you for that, Greg. So next one is from Kenneth McCabe, who is a data analyst and also an MBA candidate. And so Kenneth is wondering what data science methodologies and models you’ve used most in your career? So I guess which ones have been most useful to you?
Greg Coquillo: 23:06
Yeah. Mostly in my career I’ve seen classifications, so supervised models. Is it a cat or dog? So just being very high-level year. One of my past companies were leveraging a computer vision to determine whether the product was good or bad and the product was flying on the conveyor belt at high speed. So really classification is really so labeled data was leverage. Hopefully over the next couple of years, I’d venture a little bit more about unsupervised models. But it could be even an NLP perform different tasks there as well. So it could be something like summarization or translation, name entity recommendation. Those are the different things that I’ve seen so far.
Jon Krohn: 24:10
Nice. Very cool. Thank you for that one. And we definitely have time for a couple more here. So here’s a good one from Nikolay Kurbatov, who is an AI project manager, as well as a data scientist. Nikolay is based in Russia and he’s engaged with me a lot on the podcast. He always has great questions. So he asks what is the most common thing that you’ve experienced in terms of trade-offs for ML models push to production? So in most cases we have to choose between quality and speed. But he’s also interested if there are any particular trade-offs related to implementation?
Greg Coquillo: 24:54
Yeah, that’s an awesome question. I really liked that one. And the most common that I’ve seen is trade off in terms of risk. Typically we want to leverage AI to remove the human factor. And whether it’s human or machine, there’s always room for error, and sometimes you will have a trade-off between systems where humans tend to be a little bit more accurate, but slower versus a machine, which is faster and a little bit less accurate. And try to understand what are the risk of that lower accuracy. So being able to translate that into a business risk is the trade-off that you’d like to talk about, which are stakeholders. If machine predicted wrong, what is the risk to the business and can the business support it? And most importantly, how can the business mitigate when that error shows up? What are the processes? What are the tools? Who’s responsible for capturing this error? Who’s responsible for addressing it? And who’s responsible for continuously training these models to make sure that these errors don’t increase or improve over time?
Jon Krohn: 26:17
Right. What a great answer. I don’t know what I was expecting, but I wasn’t expecting that. And that was a great answer. I love that. And so I guess one of the things that people could be thinking about specifically perhaps following on from what you said is whether retraining their models or having the models in production to have a systems in place to be looking at for feature drift, to be looking out for inputs changing beyond the inputs, the range of inputs that we saw in training the model.
Greg Coquillo: 26:49
Yeah, absolutely. And even prior to deploying, you want to have alignment with business side as well, or parties involved in terms of what is the expected error rate would be. So say for example your precision threshold might be no more than 2%, right? Because you really care about precision for your use case. And you want to make sure you stay within there, but even aligning on that 2% is important because prior to automation, you had a higher precision or a lower threshold. Because if humans were doing it, they are doing more, they’re doing it with more accuracy and things like that. So therefore now you move into a machine which has a little bit more room to make errors. And how do you make people align on this. That when the machine will make more errors, because it’s faster that trade-off, how do you mitigate these errors? What does it means in terms of loss of business when you move precision threshold from 1% to 2%. Maybe a small number, but it’s a huge risk for a big business.
Jon Krohn: 28:00
Right. So we kind of just play around with some numbers here. So if you had a 1% error rate when humans were completely in charge of this process. So like one in 100 customers complaints. But now all of a sudden with this automated process it’s faster. It’s probably less expensive than having humans, but it has this error rate of two in 100 or one in 50. So we’re doubling our error rate. And so it’s we’re going to have twice as many complaints to our customer service center. What impact is that going to have, or we’re going to have twice as many returns. And so there’s this business needs to be prepared and you need to make sure that you’ve costed out those risks, I guess, right?
Greg Coquillo: 28:46
Exactly. Have the right processes to mitigate those things when they show up. It’s all about downstream actions when it comes to leveraging ML, AI tools and things like that. What are you doing with that prediction downstream and how does it really make your business better overall? That’s what this is about.
Jon Krohn: 29:08
Nice. Love that answer. All right. We’ve got one last audience question here. It’s from Yousef Rabi, probably Yousef Rabi and he’s a machine learning researcher. So he asks other than data improvements in terms of quality and quantity, what provides the biggest bang for the buck in terms of improvements for an AI vision system? So he’s very interested, I guess, he’s aware of your experience with machine vision projects, and yeah.
Greg Coquillo: 29:41
Yeah. So this question is quite interesting because he did hit them the best bang for the buck, right. You’re talking about a computer vision system and you’re already feeding it quality images and things like that. I think the next bang for your buck is taking a look at the whole cost of ownership of your system. Can you improve on computation power? Are you spending too much? Do you have opportunities to use less computation, use a better model that helps you compute less? Overall infrastructure, can you improve integration of systems to lower the cost of maintaining it? And then you can improve on monitoring. How are you making sure that this computer vision system is sustainable, can be upgraded very fast?
Greg Coquillo: 30:36
Does it have a monitoring bank of data that’s continuously looked at to make sure that this data is used to make it better or train a different model? Is there a really great MLOps mechanism to maintain that system? Really that’s the only thing I can see, because once you have great data, I don’t see what else. Can there be additional data that you can collect?
Jon Krohn: 31:11
Even more.
Greg Coquillo: 31:12
Yeah, because at some point, you will find that you have quality data and you’re stuck at a certain accuracy level that you cannot go past, because you can’t improve on the data anymore. What other data can you go out there and grab and label and feed into this model to make it better? Additional data collection is another way to look at it. I see three, monitoring mechanism, additional data collection, and improving the total cost of ownership. Those are the three areas that you can take a look at.
Jon Krohn: 31:54
Nice, a wonderful answer. You’ve great answers to all of these questions.
Greg Coquillo: 31:59
Appreciate it.
Jon Krohn: 32:00
Which leads me to my big series of questions. In researching for this show, I discovered that you’re developing quite a bit of an interest in quantum machine learning. I’ve seen you posting about it a lot. I see that you have a certificate from IBM on quantum computing. And so I think it’s a fascinating topic and I’d love to pick your brain about it. First of all, what is quantum computing? How is it different from traditional computing?
Greg Coquillo: 32:36
I will attempt to answer this as Greg Coquillo trying to understand this complex world of science and the curious Greg Coquillo in me, who will try to help everyone else listening to this understand. Quantum computing is the idea of exploring the world of quantum mechanics and leveraging these characteristics to build a computer that does just that, compute compared to a classical computer. When compared to a classical computer, a classical computer is equipped with bits who can be either zero or one. However, a quantum bit which is for short, cubit can be both zero and one through a phenomenon called superposition. This is something that allows quantum computing through the manufacturer of quantum circuits, which are gates that allows you to perform certain instructions called algorithms to compute them faster than normal, because you have two states at the same time, so you can reduce computational power by millions.
Greg Coquillo: 34:03
It depends on how complex the most complex factorizations, or the most complex search algorithms and things like that can be performed much faster with quantum computing. And the last thing I’d like to say is with quantum computing, which is still even though we’ve been working at it for so long seemingly, it’s still at its infancy, still no one knows what will come out of it at the end because it’s a very noisy environment. We just don’t know how things will come through, but in the end what you will find is that there will be just another way to compute on top of what you already know CPU, GPU, CPU. Now, you can start expecting QPU. And it can be focusing on specific use cases that may want much powerful computational powers. And we can talk about different industries that can leverage these, so drug development, biology to synthesize new materials, material science. Very complex use cases that requires simulations for airplane makers and things like that, that require huge amount of data, huge amount of combinations to simulate the right way to perform something, quantum computing can be helpful there.
Jon Krohn: 35:35
Cool. I did not know. It hadn’t occurred to me and I’m never going to forget. You just pointed out something to me that now is so obvious that QPUs, these quantum processing units, which are as you say, in their infancy today, these will be appended to a classical computing system. You’ll have that CPU still running the operating system on the computer and letting you look things up in stack overflow. But in the same way that today for training a deep learning model, or maybe even production with a very big machine vision model, we will have particular operations sent off to the GPU to run on that specialized processing device. In the same way, there’s going to be a relatively small number, a narrow set of tasks that we say, okay, this is perfect for quantum processing unit and we send it over there and then we get them to return the results back.
Greg Coquillo: 36:38
That’s correct. I do believe there will be something like that for sure. And also not eliminating the fact that we may also see quantum computer on a circuit board. Just like we saw with the transistors, we want to make sure we build systems with millions of cubits. I think right now the biggest system that claimed quantum supremacy has 76 cubits only. When you compare that with a chip for classical computer that has millions of transistors, we’re way out from that achievement, but we believe that it can be done. You will see that at some point as well.
Jon Krohn: 37:21
Cool. In machine learning in particular, I guess, maybe you already did give some examples of potential applications, like creating materials. We could be doing that in a machine learning driven way. When you were going through those examples, those were kind of, yeah those were machine learnings.
Greg Coquillo: 37:40
Yeah, exactly. They can be seen as machine learning, but then again, you can leverage quantum to build a quantum neural net, that adjusts the weights faster with less data, things like that. You could leverage it to build a classification model. In my particular one which is optimization use cases. The most complex flight systems or flight scheduling systems, can leverage quantum computers as a hybrid with a classical computer to optimize weights for a newer network, that is used to predict flight schedules or something like that. This is where the hybrid quantum and classical can come in very handy. One of the fallbacks for quantum computer is it’s not as precise as classical computers.
Greg Coquillo: 38:55
It’s good at giving you a probability range, a probability that this is it, but in terms of saying, this is it for sure the classical computer is better. In terms of leveraging both as a hybrid system, you can get bigger bang for your buck. And this is where I believe a lot of companies are going towards, which is leveraging a hybrid system of quantum and classical to tackle the problems now, instead of waiting five, 10 years down the road to say, oh, quantum is ready for you now to use. People are getting really smart building custom systems with these to tackle today’s problems.
Jon Krohn: 39:39
Super cool. I’ve learned a ton from that answer, Greg so I’m really glad that I asked it. I’m glad that I stumbled across this developing interest of yours. Do you have any particular recommendations of tools or resources for people to learn about quantum computing, or maybe even quantum machine learning themselves?
Greg Coquillo: 40:00
PennyLane, quantum machine leaning.
Jon Krohn: 40:02
PennyLane I’ve come across that.
Greg Coquillo: 40:04
Is a good one. I would say IBM Qiskit is another one that I can refer anybody who wants to-
Jon Krohn: 40:14
How do you spell that? Qiskit?
Greg Coquillo: 40:17
Qiskit, Q-I-S-K-I-T.
Jon Krohn: 40:22
All right. Yeah, I’m glad I have you pronouncing that for me. Nice, Qiskit. Sweet, all right, well I’m going to check those out right after the show. I get you on the SuperDataScience Show, you think you’re doing something for the audience, but really all of this was just for me.
Greg Coquillo: 40:41
Men, you’re the best man behind it.
Jon Krohn: 40:44
Because I’m giving a talk on September 10th, it’s just a 20 minute talk, but it’s an intro to quantum machine learning. I’m giving it at the machine learning conference, which is going to be in person in New York City at this cool nightclub called 235th Avenue. It’s there all day. It’s not a party in the evening, but they’ve rented out the nightclub from 8:00 AM to 6:00 PM. I’m doing this 20 minute talk an intro to quantum machine learning, and I don’t know anything about it. Almost everything I know I just learned from you.
Greg Coquillo: 41:15
You will not regret visiting PennyLane for sure.
Jon Krohn: 41:18
Nice.
Greg Coquillo: 41:19
You’ll find what you need to know.
Jon Krohn: 41:20
It was stumbling across PennyLane that led me to think, you know what? It looks so well explained here, I bet I could do a talk on this. And then I submitted the talk and when they accepted it, I was like oh no.
Greg Coquillo: 41:35
Now, you’re full of stress. Now, you truly need to rely on PennyLane, so absolutely. Let me know if you need help with that too, so I’m happy to [crosstalk 00:41:44].
Jon Krohn: 41:44
Yeah, I may need to fly you in.
Greg Coquillo: 41:47
No, from a preparation standpoint for sure.
Jon Krohn: 41:50
Awesome. I would definitely appreciate your feedback on that. Okay, sweet. All right, you’ve answered so many great questions from me and from the audience. We’ve got to learn a bit about your background as well. Do you have any books, a book recommendation for us?
Greg Coquillo: 42:07
Yeah. There’s one that I’m reading right now that I’ve been very slow digesting it, it’s called Algorithms to Live By, by Brian Christian and Tom Griffiths. And it’s really an exploration on how computer algorithms can be applied to our everyday lives. And it can help you solve common decision making problems. I’ll give you one quick thing. This book has helped me understand the process of cacheing better than anything else. And it leveraged the use case of a library that stores recently rented books in a bin at the entrance of the library, versus putting it back in its place, because these recently rented books most likely are the most popular ones. And it felt people could just use them fast as possible, instead of having to walk in the back into the sections of the library to look for it. So they cached it at the front desk for people to retrieve as fast as possible, which is an idea of cacheing in computer. So those who tends to see that what computers do we do it on our daily lives, on our daily decision making. And it’s amazing that we’re able as a human species to be creative enough, to teach a machine to do what we do on a daily basis. I think that was fascinating, so anybody can enjoy this book.
Jon Krohn: 43:49
And cacheing can be hugely useful in machine learning. With my company, we have this very computationally expensive process for taking natural language documents and converting them into this vector representation. This one dimensional vector of numbers. And that process if we were to try to run that every single time we ran our models to make a prediction, it would take days, you’d be sitting at your computer. You make a query and you’d have to wait days to get a result, but we can cache those vectors and have them readily available. And then so we can give you results in an instance as opposed to in days. This cacheing idea of the library book analogy, if people aren’t aware of cacheing, I definitely recommend learning about it either from the Algorithms to Live By book, or more generally because it can be a game changer for productionization of your models.
Greg Coquillo: 44:56
Absolutely.
Jon Krohn: 44:59
Nice. All right, Greg, we’re reaching the end of the program, but I promised at the beginning of the show, that people would be left with opportunities to get in touch with you in real time. What are the best ways? Obviously we’ve got your LinkedIn, I’ve already mentioned that so people should follow Greg on LinkedIn. That’ll be in the show notes where his profile is, if you want to follow him. And in addition to that, where can people find you?
Greg Coquillo: 45:27
Yeah, absolutely. If you want to catch me live, you’ll find me definitely every Friday around 2:30 Central Time, I believe, at the Data Science Happy Hour, hosted by the one and only Harpreet Sahota, that’s on Fridays. Then if you want to catch me on Saturdays, their Saturday is around 9:00 AM, PST. Sorry for giving you time zones all over the place. But this is with my family, the integrated machine learning, led by Dr. Thom Ives, [inaudible 00:46:03].
Jon Krohn: 46:02
I know who he is, I haven’t met him yet in person, but we’ve definitely talked.
Greg Coquillo: 46:10
You need to talk to him too.
Jon Krohn: 46:12
Yeah, all right. We should get him on the show?
Greg Coquillo: 46:15
You should get him on the show. I think this is where anybody can find me live and happy to talk to anyone. But also feel free to reach out to me on LinkedIn. It’s fairly manageable, it’s been a little bit more challenging lately because of the flood of messages and LinkedIn is not the best UI to manage messages, but I’ll try my best to respond to you at some point.
Jon Krohn: 46:39
Nice, well, that’s super kind Greg. I can’t believe how much you give to the community, but it’s certainly paying off in terms of how many followers you have in the LinkedIn Top Voices awards, so yeah.
Greg Coquillo: 46:50
Thank you.
Jon Krohn: 46:50
Just keep on trucking, Greg. So excited to watch how your career continues to develop and the huge impact that you make on people’s lives. And hopefully, we’ll have the opportunity at some point in the future, to have you back on the show and catch up with you.
Greg Coquillo: 47:05
Jon, I’d be happy to return, but thank you so much for having me for the first time ever. It’s an honor, it’s a humbling experience, and I’m looking forward to hearing more. But also most importantly, I’m looking forward to seeing you grow with this great podcast of yours. So thank you so much.
Jon Krohn: 47:22
Thank you, Greg. All right, catch you in a bit. I love how clearly and succinctly Greg answered all the questions I and the audience had for him. When we were planning the episode, I thought we had way too many topics to cover, but Greg pops in with a spot on answer, backed up with a straight forward example, and boom we’re onto the next one. In today’s episode, Greg filled us on element AI’s maturity framework for AI businesses, that AI startup success comes from understanding your long term business strategy while iterating tactically, to solve both machine learning problems and commercial problems. That while machines typically are much faster than people, they tend to be less accurate.
Jon Krohn: 48:07
When automating a business process, you need to be sure the business is prepared to mitigate, capture, or address the risks from this. That while quantum machine learning is in its infancy, some optimization problems like flight schedule prediction are likely to be revolutionized by quantum processing units in the years to come. In particular, Greg pointed us in the direction of PennyLane and IBM’s Qiskit for getting started with quantum machine learning ourselves. As always, you can get all the show notes, including the transcript for this episode, the video recording, any materials mentioned on the show and the URL for Greg’s award winning LinkedIn profile, as well as my own social media profiles at www.superdatascience.com/495. That’s www.superdatascience.com/495. If you enjoyed this episode, I’d of course greatly appreciate it if you left a review on your favorite podcasting app or on the SuperDataScience YouTube channel, where we have a video version of this episode.
Jon Krohn: 49:05
To let me know your thoughts on the episode directly, please do feel welcome to add me on LinkedIn or Twitter, and then tag me in a post to let me know your thoughts on this episode. Your feedback is invaluable for figuring out what topics we should cover next. All right, thanks to Ivana, Jaime, Mario, and JP on the SuperDataScience team for managing and producing another amazing episode today. Keep on rocking it out there folks, and I’m looking forward to enjoying another round of the SuperDataScience Podcast with you very soon.
