(upbeat music) >> From theCUBE studios in Palo Alto and Boston, connecting with thought leaders all around the world, this is theCUBE conversation. >> Hey, welcome back, everybody, Jeff Frick here with theCUBE. We're in our Palo Alto studios today. We're having a CUBE Conversation around a really interesting topic. It's applied AI, applied machine learning. You know, we hear a lot about artificial intelligence and machine learning in kind of the generic sense, but I think really, where we're going to see a lot of the activity is when that's applied to specific solutions and specific applications. And we're really excited to have our next guest. He's applying AI and machine learning in a really interesting and important space. So joining us from San Francisco is Bryton Shang. He's the founder and CEO of Aquabyte. Bryton great to see you. >> Yeah, Jeff. Great to be here. >> I can't believe it's been almost a year since we met at a Kosta Noah event. I looked it up June of last year. Wow, how time flies. But before we get into it, give everyone just kind of the quick overview of what you guys are up to at Aquabyte. >> Aquabyte's a company, we're building software to be able to help fish farmers. It's computer vision and machine learning software based on a camera that takes pictures of a fish in a fish pen, analyzes those images and helps the farmer understand the health of the fish, the weight of the fish, how much to feed and generally better manage their farms. >> It's such a great story. So for those people that haven't seen it, I encourage you to jump on the internet and look up the AWS special that Werner did on Aquabyte last year. It's a really nice piece, really gets into the technology and a lot of the fun part of the story. I really enjoyed it and you know, congratulations to you for getting featured in that AWS piece. But let's go to how did you get here? I mean, you're really interesting guy. You're a multiple company founder coming out of Princeton, in most of your startup role, your startups are all about, Applied Mathematics and Statistics but you've been in everything from finance and trading to looking at cells in the context of Cancer. How did you get to Aquabyte? Was it the technology? And then you found a cool solution? Or did you hear about, you know, an interesting problem and you thought, you know, I have just the trick to help attack that problem. >> Well, so I had studied Operations Research and Financial Engineering at Princeton, which I guess we would call nowadays, like modern day machine learning and data science. So that was something as you mentioned, first I'd apply it to algorithmic trading, and then got on to more general applications of computer vision for example, in cancer detection. The idea to apply machine learning talk to aquaculture, came from a number of different sources. One was from a previous co-founder who had been doing some investigation in the fish farming space, had a business school classmate who owned a fish farm. And also growing up in Ithaca, New York near to Cornell I had a family friend who is a professor of aquaculture. And really just to learn about fish farming and overfishing and the idea that over half the fish we eat nowadays are coming from fish farms and that you could use machine learning and computer vision to make these farms more efficient. That being very interesting and compelling. >> So it's really interesting. One of the things that jumped out from me when I watched the piece with Werner was the amazing efficiency on the feed to protein output in fish farming. I had no idea that it was so high, it's basically approaching one to one really interesting opportunity. And I had no idea to that, as you said over 50% of the world's seafood that's consumed was commercially farmed. So really a giant opportunity and so great space to be in a lot of environmental impacts. So but how did you decide to find an entree? We know where to find an entree for machine learning to make a big impact in this industry. >> So it came from a couple different angles. First, there's been applications of machine learning computer vision and other industries that served as good parallels where we're using cameras to be able to take images and then use computer vision to derive insight from those images. For example, just take aquaculture where you're using cameras to spray weeds to understand crop yield. And so there's good parallels and other industries. aquaculture specifically, I was also looking at what was coming out in the machine learning literature in terms of using cameras to size fish. And so the idea that you could use cameras to size fish was very interesting because then you can use that to figure out growth rates and feeding. And as I developed my idea, it really became clear that you could use computer vision and machine learning to do a wide range of things at the farm and so, it started with this idea about using cameras to size fish and then it became monitoring health and sea lice and parasites and then ultimately, all the aspects of the farm that you would want to manage. >> And correct me for wrong, but do you guys identify individual fish within the population within that big net and then you're basically tracking individuals and then aggregating that to see the health of the whole population. >> That's right, the spot pattern on the fish is unique and we have an algorithm that's able to use that to determine each individual fish via the spot pattern. >> Wow. And then how long once, once you kind of got together with the farmers to really start to say, wow, we can use this application for, as you said, worrying about lice and disease control and oh wow, we can use this application to measure growth. So now we know the health of the environment or wow, now we know the size so we can impact our harvest depending on what our customers are looking for. I assume there's all kinds of ways you can slice and dice the data that comes out of the system into actual information that can be applied in lots of different ways. >> Right So I started the company back in 2017. And if you think about aquaculture, it's actually a hugely international industry 99% outside the US, and within aquaculture, very quickly zeroed in on salmon farming, and specifically salmon farming in Norway. Norway produces about half of the world's farmed salmon and ended up going there for a conference Aqua Nor August of 2017 and whilst there had my idea and a prototype for sizing the fish with a camera, but then also realized in Norway they have recently passed regulations around counting sea lice on the fish so this is parasite that attaches to the fish and is regulated and pretty much every country that grows fish in the ocean and farmers asked me then, okay, if you could use the camera to size fish, can you also count sea lice? And can you also detect the appetite? And then it just turned into this more platform approach where this single camera could do a wide variety of application. >> That's awesome. And I'm just curious to get your take on, the acceptance and really the excitement around, you know, kind of application of machine learning in this computer vision in terms of the digital transformation of commercial fish farming, because once it sounds like once they discovered the power of this thing, they very quickly saw lots of different applications, and I assume continue to see kind of new applications to apply this to transform their business. >> Right, I would say fish farming itself is already fairly highly mechanized. So you're dealing with fairly rough conditions in the ocean. And a lot of the equipment there is already mechanized. So you have automatic feeders, you have feeding systems. That said, there isn't too much computer vision machine learning in the industry. Today, a lot of that is fairly new to the farmers. That said they were open to trying out the technology, especially when it helps save labor at the farm. And it's something that they have familiarity with, with some of the applications for example, with Tesla with their autopilot and other examples that you could point to in common day use. >> That's interesting that you brought up Tesla, I was going to say that the Tesla had an autonomous driving day presentation. I don't know, it's probably been a year or so now but really long in-depth presentations by some of his key technical people around the microprocessor and AI and machine learning and a whole thing about computer vision. And, you know, there's this great debate about, can you can you have an autonomous car without Lidar and I love the great quote from that thing was you "Lions don't have Lidar "and they chase down gazelles all day long." So, we can do a lot with our vision. I'm curious, some of the specific challenges within working in your environment within working in water and working with all kinds of crazy light conditions. It's funny on that Tesla, they talked about really some of the more challenging environments being like a tunnel, inside of a tunnel with wet pavement. So, kind of reflections and these kind of metric conditions that make it much harder. What are some of the special challenges you guys had to overcome? And how much, is it really the technology? Or is it really being done in the software and the algorithms and the analyzing or is it basically a bunch of pixel dots? >> Right. The basic technology is based on similar, it's a serial camera that takes images of the fish. Now, a lot of the special challenges we deal with relate to the underwater domain. So underwater, you're dealing with a rough environment, there could be particles in the water, specularity some reflections underwater, you're dealing with practical challenges such as algae, but even the behavior of the fish, are they swimming by the camera? Or do you want to position your camera in the pen. Also, water itself has interesting optical properties. So the deeper you go, it affects the wavelength that's hitting the camera. And also you have specialized optics where the focal length and other aspects of the optics are affected underwater. And so a lot of the specific expertise we've developed is understanding how to sense properly underwater. Some of that is handled by the mechanical design. A lot of it is also handled by the software, where on the camera we have GPUs that are processing the images and using deep learning computer vision algorithms to identify fish parts and sea lice and other aspects of the fish. >> It's crazy, and how many fish are in one you know, individuals are in one of these nets. >> So single pen can have as much as 100,000. Where actually in one pen, which is I think it's the largest salmon farm in Norway based on an oil rig called the ocean farm where they have 2 million fish in a single pen. >> 2 million fish, and you're in that one. >> Right, yes. >> And you've identified all 2 million fish or do you work on some sampling? Or how do you make sure every fish eventually swims by the camera? Or does the camera move around inside that population? That's an amazing amount of fish. >> So I think we'll eventually get to the point where we can identify every single fish in the pen and use that to track individual health and growth. Well we practice what we use the individual recognition algorithm the deal is to de-duplicate fish. So a common question we get asked is okay, what if the same fish swims by the camera twice, and so it's used to de-duplicate fish But I think eventually you'd be able to survey the entire population. >> That's crazy. So where do you guys go next Bryton, again you've brought your analytical brain to a number of problems. Do you see kind of expanding the use within the fish industry and kind of a vertical player? Do you see really a horizontal play in different parts of agriculture and beyond to apply some of the techniques and the IP that you guys have built up so far? >> Well, starting with Norwegian salmon, we want to bring this to other countries around the world for other species. So we've expanded to our second species, which is a rainbow trout. We also are, starting with computer vision are building this very interesting data set which we can use to enable other applications. Eventually, we'll get to the point where that data allows us to run fully autonomous fish farms. Right now the limitations of fish farming is that it needs to be close to the shore. So you can have people go to the farms. And once you have fully autonomous fish farms, then you can have fish farms in the open ocean, fish farms on land. And with the world being 70% water, we're only producing about 5% of the protein from the oceans. And so it presents a massive opportunity for us to be able to increase the amount of world's demand for protein. Also given that we're running out of land to grow crops. >> Wow, that's amazing. We're only getting 5% of our food protein out of the ocean at this stage? >> Right, right. >> That is crazy. I thought it would be much higher than that. Well, certainly a really cool opportunity and, a kind of a really awesome little documentary by Werner and the team, definitely go watch it if you haven't seen it. So I just give you the last word as you've been in this industry and really seen kind of the transformative potential of something like computer vision in commercial fishing and who would have even thought that, six or seven years ago? How does that help you kind of think forward, kind of the opportunity really to use these types of applications like computer vision and machine learning to advance something so important, like food creation for our world. >> I think there's definitely a lot of opportunities to be able to use machine learning computer vision, similar technologies to help make these industries a lot more efficient. Also a lot more environmentally sustainable. I'd say something like this industry, like aquaculture, it's not so apparent just if you're in the valley, and even in the US just because 99% of it happens outside the US and so to be able to be familiar with the industry to know that it exists and to build applications itself is a bit of a challenge. I would say that is changing. One of the things that actually came out a couple weeks ago was an executive order to actually start kick starting offshore aquaculture in the US. So it is starting in the US. But more generally, I do think there's a massive opportunity to be able to apply machine and computer vision in new industries that previously haven't been addressed. >> Yeah, that's great. And I just love how you got kind of a single source of data, but really the information that you can apply and the applications you can apply are actually quite broad. It's a super use case. Well, Bryton, thanks for spending a few minutes. I've really enjoyed the story. Congratulations on your funding rounds and your continued success. >> Thanks, and really appreciate to be on and yeah, hope to continue to help bring the world more sustainable seafood. >> Absolutely. Well, thanks a lot Bryton. So he's Bryton and I'm Jeff. You're watching theCUBE. We'll see you next time, thanks for watching. (upbeat music)