Machine Learning on the Edge Episode 3 — Guido Jouret on Data-Enabled Industry

In this installment of Machine Learning on the Edge, Zach Shelby, Edge Impulse CEO and co-founder, and Guido Jouret, veteran technology leader (Cisco, ABB), board member (Polycom), and startup/VC advisor, discuss how IoT innovation is impacting industry thanks to advancements in AI and edge computing, how applying data to different industries has proved challenging, and more on connectivity, processing power, and software innovations.

We’ve excerpted some abridged moments from the conversation; watch the full conversation here, or listen as a podcast.


ZACH: We both came from the space of IoT innovation. Almost a decade ago, I was at Sensing Node working on electric meters, street lighting, enterprise-scale wide area networks, and then we were acquired by Arm. At the same time, Cisco was investing big in IoT. Where has IoT come since then?

GUIDO: I think there’s really sort of two IoT markets that have emerged. One is the consumer side, and we know this through the connected home speakers, the smart light bulbs. You can change the colors and things like this. And sadly enough, there’s been a proliferation in the consumer side, but no real compelling use case. It’s entertaining to be able to do that. But it’s not changing your life. On the industrial side, there’s a lot more that is yet to come. Smart meters. We’re going to see more and more autonomous vehicles. We’re going to see more and more innovations in the health care sector, for example wellness all the way towards patient care. That’s still sort of a work in progress. But one of the reasons why the progress has been slower, I suspect, than both of us may have wanted, is that it’s a really hard problem. You have to connect things, get the data, compute on the data, interpret the data and take action. And we’re probably going to dig into this some more. It’s a really hard problem, but a very worthwhile problem to solve.


ZACH: When you understand machine learning deeply enough, you can come to terms with its probabilistic nature. You can apply that to solve problems in safe ways. But the rest of the engineering world is still far behind that.

GUIDO: Machine learning model is a black box. You put a whole bunch of inputs together, you train it on the desired outputs and you try to fit the model that you want. But ultimately, especially with mission critical health, safety, and really important kinds of automation jobs, you’re still struggling with the idea of, “well, how will this behave at the boundary conditions? What if I present it with some data or a condition that was not in my training set?” If it’s just a technology that you use to generate some pretty pictures on, say, Midjourney or some other application, you get something that looks like it was cooked in the microwave. Not a big problem. But if this is now controlling a vessel or somebody’s insulin pump or it’s being used to control the energy in a factory, the repercussions could be pretty severe.


ZACH: I like when we’ve talked in the past about physics-based models. Are we going towards a world of hybrids and boundary conditions?

GUIDO: One of the reasons we want to go to ML is because the real world is so multidimensional and so complex. The physics models by themselves will not get us there. We also need to be able to insert the human in the loop. So when the model has a low confidence prediction, it says, “Okay, I’ll give you an outcome, but I didn’t have a lot of training data on this. So therefore you take this caveat emptor, you know, take this with a grain of salt.” How do we seamlessly then transition to a human operator who can take over and ideally teach the model to say, “next time this happens, do it like that?”


ZACH: You’ve done a lot of work in digital health, medical, pharma. How do you see this evolution of machine learning and data being applied in this space?

GUIDO: I think that health and wellness is probably the biggest potential market for IoT of all. Here in the United States, health care is really 19% of GDP. We spend almost more than double what most OECD countries spend on health. When I was at Nokia investing in this area, our team got together and said, “how do we look at this market?” We thought of it in terms of at least two, but there’s probably a third market in between as well. The first market is really what we call the worried well. Instead of making you wait until you get sick, how do we make sure that you stay healthy? The Oura ring is an example of that. Sleep activity, mood, there are a lot of different things you can infer. How do we use the data that we have from those devices to nudge you into good behaviors? That’s the first problem. Then at the other extreme, you have remote patient care. Devices that are monitoring stroke victims, for example, or inside the hospital looking at the patient’s EKG and temperature and heart rate. And that is all about trying to provide continuous data and feedback on treatments that are working and how well they’re working. But I believe there’s a giant market in between which we can go into as well that could be sort of a stepping stone. How do we go from the unregulated or lightly regulated wellness market towards this sort of much more difficult and challenging medical market? A lot of people see that as a binary thing, and I actually think that it’s a continuum.

ZACH: I’m already seeing machine learning applications on the very beginning of medical devices. And so, passive monitoring of things like glucose, there’s so much that we can do and that are within limits of what the FDA will let us do today. Then there’s whole areas of medical that are still pretty much off limits because the FDA hasn’t really decided what it thinks about machine learning. It’s kind of left as an open question at the moment. And so there’s a lot of hesitation there to apply machine learning. But what other markets can we go after in the meantime?

GUIDO: One area is the whole pet care market. For dogs, cats, cows, there’s a whole bunch of applications there where the hurdle of proof and the regulatory framework is not as severe, understandably, as it is when you’re experimenting on people. Another market where you’re not experimenting on people, but where you can apply the same wellness technologies to actually affect a much more important outcome, I think, which is to take care of the elderly. When, most of the time, you ask, “Hey, how do you want to live your golden years?” They don’t want to be living in an assisted care facility. They ideally want to stay in their homes, but the people that look after them or that care about them, they worry. And they worry because they say, “What if you fall down? What if you know you’re not getting enough activity?” So a lot of these technologies develop for the wellness market with different algorithms that can detect, for example, a difference in lifestyle. You’re getting up later, you’re not moving around as much. Your gait has changed. So how you walk has started to change, and doctors have correlated the degradation in gait to life expectancy, limitations and all the rest. There’s a lot of really interesting applications of that technology that can be put in place to allow people to age better in place and to enjoy their golden years.

ZACH: A lot of what we’ve been talking about, those innovations, this new application of machine learning and data and IoT, these are going to be driven by startups. How do you encourage the next generation of startup entrepreneurs now to go take advantage of the current market? And what kind of product should they be building in IoT and machine learning? Where does it make sense to go succeed?

GUIDO: I look at it in terms of supply and demand side. All of a sudden this crisis, the economic recession that’s happening, is a boon on the supply side because although people are hit by layoffs and of course a lot of companies are struggling to come up with cash, if you do have the resources, i.e. the cash, finding good talent has all of a sudden become a lot easier. So all of a sudden, people who might not have considered working at a startup maybe don’t have a choice, or they’re saying, You know what, that sounds pretty good right now, so that’s great for you. Also, things like real estate has now become cheaper because buildings are lying empty. So I think that creates a great sort of possibility of supply.

On the demand side, what recessions and important events like this tend to do is they cause a mindset shift. Basically people think differently about doing something in a way that maybe they hadn’t thought of before. During the pandemic, with people working from home, people had said, “No, that’ll never work.” And yet in the case of, say, telemedicine adoption, now post-pandemic is 40 times what it was pre-pandemic. All of a sudden there’s this huge shift in an acceptance of something. And when that happens, that creates demand. For a startup, you have supply and you have demand. And these tectonic shifts that happen, in moments like these, are usually when the next great companies are born.

Subscribe

Are you interested in bringing machine learning intelligence to your devices? We're happy to help.

Subscribe to our newsletter