Bear Necessities: How Technology is Changing Wildlife Research in Ecuador

Brian: Hello, and welcome to the Arm Viewpoints podcast, where we explore technology topics at the intersection of AI and human imagination. I’m your host, Brian Fuller, Editor in Chief at ARM. In this episode, we continue our exploration of how AI is revolutionizing wildlife conservation. This time we’ll focus on the elusive Andean bear, one of South America’s most iconic species.

Joining me are two guests. At the forefront of this exciting work, Becky Zug, director of the Carnivore Lab and professor at Universidad San Francisco de Quito, who has dedicated her career to studying and protecting Andean bears, and Ed Miller, a senior principal engineer at ARM and co-director of the Bear ID Project.

It was applying cutting edge AI technology for wildlife conservation. In our conversation, we’ll touch on the unique challenges of studying and protecting Andean bears in Ecuador’s fragmented habitats, how AI powered individual identification is transforming bear research and conservation efforts, the potential for this technology to unify data.

Disparate research projects across South America, the role of arm-based hardware, processing vast amounts of camera trap data, how this work could be extended to other elusive species like pumas, the broader impact of digital technology on biological research and challenging environments. And much, much more without further delay, I bring you Becky Zug and Ed Miller.

So, Becky. Ed, thanks for joining us today. Let’s start off with just some quick introductions. Becky let’s start with you. Give us a little bit about your background, how you got to where you are today, and then we’ll throw it over to Ed.

Rebecca Zug: Okay, great. Thank you so, much for having me today. I’m excited to talk about our work and this project and BEARS in general.

So, a little bit about me. I’m American, but I’ve been living here in Ecuador for quite some time now, I think maybe 15 years or so. But I started working down here in 2007 as a graduate student. I got into Andean bears because I was interested in coming to South America. It looked like my life was heading down here.

And I was very interested in carnivores and their conservation, but also the issues they cause when they live in human dominated landscapes. So, that’s really how I started looking toward Andy and Bears here in Ecuador, but I actually have an undergraduate background. I have a BFA and I spent a summer working at National Geographic in their photo promotions department.

And that really got me thinking about ecology and sciences much more than arts. This led to a semester in East Africa and that kind of did it for me and I slowly moved away from the arts and into the sciences. So, long story short, that’s how I ended up here. And right now, I’m a professor at a university here in Ecuador and the director of the carnivore lab here.

Brian: You just preempted my next question for you, which was how did you go from a bachelor’s in photography at Syracuse to diving into the sciences at Wisconsin, but you just answered that.

Rebecca Zug: Yeah. Yeah, I know it’s a big jump and I think we often think of the sciences and the arts as polar opposite, but I think there’s an investigation and a Creativity that comes with both, so, I had to do quite a lot of work to go from a BFA Into a science graduate program by taking classes and doing fieldwork.

I spent some time in Southeast Asia Really figuring out what I wanted to do. Was it protection? Was it enforcement? Or was it research? That’s where I’ve ended up with. It was a bit of a journey between art and sciences, but it was, I would say it’s that, that study abroad program in East Africa where I really got to be around a whole lot of wildlife and got really excited about understanding them and conservation.

Brian: Yeah. Those programs are worth their weight in gold to help develop younger people. Okay, Ed, over to you. You have an equally interesting story about how you bridge the technology environmental divide.

Ed Miller: Yeah. Thanks, Brian, for having us again. Yeah, I’m a senior principal engineer at ARM. But I also am a co-director and volunteer developer at the bear ID project. My background is, I have a, an engineering degree from Carnegie Mellon university. I did some hardware development in early parts of my career, software and software management later on in my career. And also, I’ve got a passion for wildlife and the environment. Which kind of led me down this journey of combining machine learning and my passion for wildlife into What’s now become the bear ID project.

Brian: So, Becky let’s start at the top with, and I’m going to butcher this, Tremarctos ornatus, the Andean bear. What why bears? Why Andean bears? Let’s start at that level.

Rebecca Zug: For me, why Andean bears? Ands, big carnivores and species that have similar characteristics like chimpanzees or elephants even, have very similar conservation challenges, right?

They can be big, they can be dangerous, they can be difficult to live next to. They need a wide variety of resources sometimes; they need huge spaces to get all of the resources they need. And so, as we grow as a human population into wild places, there’s less and less space for these species.

Ands, that, that tension line between having these animals in these wild spaces, but also being able to support the human communities that need resources is of great interest to me. So, not only their ecology is interesting to me, they’re the only bear in South America but also finding a place for them in, in.

Crowded places like Ecuador are also a conservation puzzle and they’re not unique in that way. And I think of the solutions that we have for Andean bears. We can apply to other species like pumas and jaguars and of course other species of bears too. And so, the evolutionary history of the species is of interest.

And the fact that they’re a unique bear. But also, the solutions to keeping them for the long-term in the wild in healthy populations is, has always been of great interest.

Brian: And there, I understand there are an umbrella species, which, conjures up interesting images of bears walking around with being protected by the rain, but that’s not what it is.

Tell us what are umbrella species?

Rebecca Zug: That’s a great question. Ands, that’s a, an important conservation term. So, we, we can’t. Measure everything we can, protect everything we can, get people as interested and our smaller species plants, right? Moths and frogs don’t get the attention that bears do.

But because bears use a variety of habitats to collect the food resources they need, and because they need a large amount of space, if we protect Andean bears, The idea is that the Andean bear itself is holding an umbrella of protection for the other species it shares the landscape with. Ands, that’s where we get that term from.

One of my favorite pictures of this is an Asiatic black bear holding an umbrella. I think holding hands with a tiger and they’re both holding the umbrella over pangolins and other smaller species that get less conservation attention. So, the idea is that if we can protect Andean bears, we protect the other species they share.

They share the space with.

Brian: Is it, does that have to do with the philosophy of species protection or the more you protect an umbrella species, the more beneficial it is in the trickle-down effect throughout the food chain?

Rebecca Zug: It’s really a conservation term. It’s a conservation approach, right? And so, if we can gain enough attention to protect the Páramo habitats, the high-altitude grasslands where white tailed deer.

Many endemic plants live as well Andean foxes and populations of pumas and mountain tapirs. If we can protect the Páramo and the cloud forest where Andean bears also live then we protect the species that also live in the cloud forest, which is a whole different group of species. So, it’s a conservation term, but it ends up being a trickling down ecological, ecologically, if you if you want to put it that way.

But if we put aside areas and say, okay, Andean bears live here, we’re going to enforce the borders. We’re going to make sure there’s no encroachment. We’re not going to convert this land to agriculture, then the other species benefit from that protection of bears. And so, usually an umbrella species is one that uses a variety of resources, needs a whole lot of space, gets a whole lot of conservation attention.

Just like a bear mite or an orangutan or even elephants or some of these big charismatic megafauna, often we use them is this idea of umbrella species. And Andean bears ecologically as well. Play an important role in those ecosystems. They’re not quite engineers, right? But we think that they’re seed dispersers.

So, they help the forest. They certainly participate in the nutrient cycle. They move things around. They break trees, right? And so, they do have their important ecological role. So, if we maintain species, it’s also helps the health of that ecosystem. And then ultimately, human health, right? These ecosystems provide us with incredible ecosystem services.

Like water and fresh air and food and medicine and shelter and all these different things. And Andean bears can be a nice conservation surrogate for that kind of protection.

Brian: So, I’ll throw this to, to both of you, given your involvement in this. And Ed, in the other episode with Melanie Clapham, we talked about her work up in Canada.

How are, and we’ll talk about the facial the facial recognition differentiation in bears, but how are Andean bears different from other bears? Not only physically, but in their habits and how they affect their habitat.

Ed Miller: So, certainly from the technology perspective, I would say the biggest difference is, That the Andean bears are a lot more elusive than the brown bears in the habitats that we’re studying. When we go through camera chop footage from Melanie’s bears in British Columbia, the brown bears there we get a large hit rate of brown bears.

There are other animals in the area, but when she’s putting things near a salmon spawning ground, for example, there’s a high hit rate of brown bears. And so, that gives us a lot of data to train with and to work with. In the camera trap data that Becky has that’s not always the case.

The Andean bears first of all, aren’t just happily using these well marked bear trails that the brown bears are using and giving us these beautiful facial and profile images. And also, they’re not necessarily the most common animal on those. On those camera trap data. So, Becky, I don’t know if you want to add anything.

So, they’re shy is what you’re saying.

Rebecca Zug: Exactly. Exactly. They’re shy. And we see the kind of data that the polar bear and the brown bear people have and the black bear, the American black bear people and the, and bear biologists were blown away by the sample sizes and how you can really see their faces and you can just sit.

along a stream or at least the camera can, and you can watch multiple individuals together. So, Andy and bears are elusive. They’re shy. They’re hard to get close to and without camera traps, right? Before we had camera traps, maybe 20 years ago when they were, they became cheaper and more easily available and widely used, we didn’t know a whole lot of basic information about Andy and bears.

So, camera traps have really changed how we study this really shy species. But still, to get a good facial picture of an Andean bear, it can take some skill and some time. And then in certain ecosystems, they seem to occur to quite a low density. So, it takes quite a lot of patience as well. We don’t typically have many places we can go to where Andean bears congregate.

So, I say that just as the season here in Ecuador has started in one of our cloud forests where there’s a small avocado like fruit that, that is coming into season and the bears will come down and come from across the landscape and then climb these trees. And it’s one of the only places where you can see Andean bears together and take pictures of them and tourists can go visit them.

But this isn’t typical for the species, usually they occur in the cloud forest they’re hard to see in the cloud forest, they’re even hard to camera trap, and then if you’re up in the grasslands, usually you can’t get very close to them, so, they’re still quite at a distance. They’re a tricky to study species at a low density, a shy, generally a shy personality but camera traps and long-term camera trapping projects Has changed that for us.

It’s a, it’s given us so, much data. It takes a lot of patience and time. So, the data set, when Ed first told me how many pictures are needed to identify an individual, I was really, I was like we’re never going to get anywhere close to that, but I think there’s lots to work with here and we’re hopefully going to be able to access other data sets as well to build our sample size for the work that, that Ed and the Bear ID project are going to do with us.

So, they’re a tricky species, but we’re finding a way.

Brian: It’s gotta be a little easier. I read one of your papers from 10 years ago now, and that was about human identification of these bears and what the false positive rates were and all that sort of stuff. Let’s use that as a point to dive into this technology a little bit.

How did this, how did the bear trap project start, Ed? And how does. How did the process of developing the identification models, training sets, and so, on, how did that evolve?

Ed Miller: Becky can certainly talk to more of the camera trapping piece of it. But in terms of how we’re utilizing the data we’ve alluded to these face markings that are on the Andean bears.

This is why a lot of people might know them as spectacled bears, because they have these. White patterns around their eyes and onto their chests. And these patterns are unique, but they’re not always so, visible depending on the lighting and in the individuals, they could be more or less present, but this does give us something to grasp onto where machine learning based applications could also take advantage of these same markings.

And we had done the Bear ID project did a project in collaboration with the San Diego Zoo Wildlife Alliance. And we did a trial of Bear ID on Andean bears using images of bears from zoos. So, this was a rather small data set compared to what we had for the brown bears. But even though it was a small data set, the results were quite promising, we think, because They have these unique markings now that challenge comes a little bit even more so, with the wild bears because we’re again not necessarily getting the same facial shots and we have a lot more challenging lighting conditions and so, on, but we think that the success rate is we think we have a pretty good chance for success because these markings can help us build.

Useful data sets with even with smaller numbers of images. How did you

Brian: two get connected to begin the project?

Ed Miller: So, it was actually through the San Diego Zoo Wildlife Alliance. One of the main collaborator who worked with there was, is Russ Van Horn. He’s the zoo, the San Diego Zoo Wildlife Alliance representative for Andean bears in the bear community.

And we had gone to him, my partner and I had gone to him about wanting to extend the Bear ID project beyond the brown bears and wanted to work with a researcher that was working with them in the wild. And he connected us with Becky, who he has also done some work with. And yeah, we had a few initial discussions and thought, yes let’s do this.

Brian: And that’s the San Diego connection. I noticed that in some of your papers, Becky you collaborate with a professor from San Diego. So, that must be the.

Rebecca Zug: Yes, that’s exactly right. So, Russ Van Horn is someone who I met when, I think during my master’s program. So, quite a while ago we initially got in contact and then he’s the lead author on the paper that, that you referred to.

And seems to be someone who brings people together who have similar interests. So, he first, as Ed said, got us in touch but then of course has worked with Ed and Bear ID on other aspects of this project. But yeah, so, that’s how we got in touch was through Russ and his ability to connect people.

Brian: So, Ed comes to you and says, hey, got this idea. Were you like. This is the answer to all our problems, or I can’t imagine how we’re going to make all that technology work in a way that gives us useful data. What were your initial impressions?

Rebecca Zug: I was, it took me a while to realize what he was offering, because it was amazing.

It’s what we’ve been talking about, what we’ve wanted for years. Often in the camera traps, we have pictures of spotted cats and so, we’ve talked about this wouldn’t it be great to have AI to be able to do this, when most of us are working on a shoestring budget with way too much to do anyway.

And so, as a biologist. pursuing something like this was a bit outside my scope. And so, when Ed suggested this and I realized that they were willing to come to Ecuador to spend days and weeks with me to talk about this, to develop this tool I was blown away and thrilled. So, it was like this incredible, exciting gift.

From the tech world to help the conservation world and having, I knew about his work with Melanie and the Bear ID project and to be able to apply it to this species was, I was blown away by the offer. So, I was thrilled to say the least, I was thrilled. At the idea of the collaboration.

Brian: And he did it on his sabbatical, no less. He could have been sitting on a beach somewhere, but no. This is what I

Rebecca Zug: said, this is what I said too, but no, he came down here.

Brian: So, let’s talk a little bit more about the technology. Ed, talk us through I think this project used an arm-based Ampere. computer, talk about the hardware as well as the software side of the equation.

Ed Miller: Sure. Yeah. We talked about the camera trapping and I, Becky alluded to how camera trapping has enabled a lot of research that wasn’t possible before, but really a lot of time is spent manually processing that data. So, looking through these camera traps, looking. First of all, to see if there was even an animal, because maybe it was triggered by leaves or lighting change or whatever.

And then, if there’s an animal, you need to identify which species it is. And then when you get down to the species that you’re studying, like the Andean bear, now you have to go and figure out who that is. And I think Becky can correct me if I’m wrong, she probably ends up spending more time doing that than actually answering the research questions that she’s really, she’s really there to there to care about.

I had a lot of this is done. So, a lot of this is done manually, just viewing it on PCs in the lab. And as she said, on a shoestring budget, those PCs aren’t exactly the latest and greatest. And, trying to run AI on them is a little bit challenging. So, what we’re aiming to do is build a local server.

So, somewhere where they can keep all their data resident within the university, or within the country, initially. That’s what we’re there are some web-based solutions, but they’re mostly all based in the U. S. or Europe and require that you share all your data and, ultimately, maybe that’s a good thing to do.

But initially, the aim is to have something that can run locally, something where there’s a centralized database, starting with Becky’s lab, the carnivore lab, but then hopefully extending into other projects across the university. In the cloud forests in in the Amazon basin and so, on. So, this required a server. So, the university has servers. They have a computer science department that runs servers, but the biology team doesn’t have their own server. So, we worked with Ampere Computing, who build server grade chips based on ARM architecture. You can get them in web services like Oracle and others today.

And you can also get them as standalone servers. So, they found out about the project. We’re very interested in working with us and donated one of their servers to the Bear ID project, which we in turn donated to the university. And the intention is to use this server to, first help process the data using, automated AI technologies, but then also to maintain a database of all these observations.

that can be used across research projects here and into the future. Whereas today you might have to go dig through, a hard drive that’s been sitting on a shelf looking for a spreadsheet that has some information about a project that you did in 2017. And it’s, it, the intention here is to have a nice centralized application for doing this working with an open software application called Trapper, which was built for exactly this kind of purpose.

And so, this open source. Technology and this highly optimized efficient arm computing gives us a good combination for building a low power centralized server, which then has not only all these applications for research, but because it’s arm technology and it’s highly optimized by Ampere to run on their servers.

It’s also a very. Low power and more environmental solution for providing this kind of technology. So,

Brian: now Becky you’ve had the mind blown moment when you under, you understood the potential of this technology and now take us back to the initial questions that you wanted to explore or the themes that you wanted to explore as the project began, what were you looking for?

Rebecca Zug: Yeah. Do you mean, just to clarify the question specifically with the individual ID capabilities of the software that we’re going to work on together?

Brian: Yeah. What did you, once the data started to come in, what sort of questions were you looking to answer?

Rebecca Zug: We’re still very early in the process with Andy and Bear’s trying to get the get the software away from.

Focusing on brown bears into this species, because the, their facial pattern, as Ed said is such a big tool but part of, and again, what Ed said is that this process is so, time consuming, and you have one person going through, is bear A the same as bear B? Okay, it’s not, so, bear A and B are different are they the same?

Is one of them bear C and then D, and as you get a bigger and bigger population, you end up with this matrix of. Of lots of decision making and investment of time. And so, part of the motivation is saving time and just as I’d said again, getting into the research questions. So, what I’ve always wanted to do is that there’s a handful of us here in Ecuador working on Andean bears and we have all these hard drives sitting on all these computers across the country and after your bear project is done or you’ve submitted your report, lots of time at times that just sits there.

So, one of the neat things that I would like to do in the future is be able to bring all these the, this data on bears together, almost in a bear Facebook, if you will, that we’ve seen used for other species, like lions, like African lions in East Africa, where we know if what happened to the bear X, we haven’t seen him for three years.

Did he go somewhere else? Did he disperse somewhere else? Maybe someone has seen him in their camera traps. And if we can unify some of this information on this server and make it available to all bear researchers, then we can hopefully understand some of these bigger movement questions as Ecuador experiences heavy rates of habitat fragmentation.

Andean bears across their range are threatened by habitat loss. But also, habitat fragmentation and a bear walking across private lands past cows through pastures is really at quite a risk. So, we don’t know how fragmented these populations are and we don’t know if bears are moving between big distances of populations.

We’re doing more with GPS collars, right? Right now, but this is a fairly invasive activity. If we can use individually ID bears and have some collection of information for the, I don’t know, the keto population of bears up in the highlands and how that goes down into the Amazon. And are these the same bears?

I think this is this incredible tool. That’ll allow us to use much bigger data sets and pull together all of these bubbles of research going into something bigger for bear conservation and reconnecting these landscapes. So, that’s one thought that I had in terms of being able to use the software to not only make the process goo, much faster but unify these data sets that are often just sitting there independently unused after a project finishes.

And I’ve got lots of other ideas, too. There’s lots of places to go.

Brian: We’ll get to those for sure. So, I think earlier in the conversation, you referenced something about a lack of, for lack of a better phrase, elbow room between the bears and the population in the country of Ecuador. Can you give us a little bit more context?

What’s the land mass of Ecuador and can you put in the context the

Rebecca Zug: Yeah. Okay. Great. Absolutely. So, Ecuador is about the size of Colorado. We have the Andes run down the middle of Ecuador. And when the Andes came 20 million years ago it changed everything. And suddenly add this huge uplift of mountains, separating the Amazon from the Choco, this other rainforest on the Western side.

And we have all of this, these altitudinal distribution of species. And so, Andean bears are really an Andean species. Yes. They occur typically at higher altitudes even though we’re finding more and more bear populations or remnant populations at lower altitude, more in Peru than what we see here.

But sometimes we see overlap between Amazonian and Andean species and Andean bears. And so, we’re just learning really the extent of their distribution. But typically, if you look at their, the population distribution in South America, it follows the Andes really accurately. But as a mountain species they occur in kind of a fragmented mountain landscape.

And here in Ecuador, are most of our bears from our understanding now occur in higher altitudes. where we’ve had occupation of humans for a very long period of time. They can use a variety of ecosystems. Like I was saying that the Páramos or the high-altitude grassland, but also montane forest, and then lower-level cloud forests, but we also have the densest population of people in South America and the highest rate of fragmentation.

Or a highest rate of deforestation in South America and Ecuador has had that unfortunate honor for decades. And so, we’re losing a huge amount of forested space every year. So, meaning that people and bears are closer and closer together. And so, habitat loss, fragmentation, but then also conflict with humans is a massive threat to bears here in Ecuador.

So, even though Andean bears are second only to pandas in terms of their herbivorous diet they still have this tiny part of their diet where they can potentially eat animal flesh. And of course they eat insects, right? And over 300 species of plants, but they can also, they’re big and they can attack and kill livestock.

And if you are a rural landowner and you have three or four cows and a bear comes and kills one of your cows, that’s like someone reaching into your bank account and taking a quarter or half or more of your savings. So, it can be devastating to have a bear kill livestock to some of these rural communities and landowners.

And the reaction tends to be quite not necessarily aggressive, but quite strong. And you can’t, you can’t blame them for wanting to protect their livelihood. So, working with landowners, trying to keep bears away from cows and cows out of bear habitat is an important part of conserving them here in Ecuador.

And so, the deforestation and the dense population and the people inside bear habitat with bears crossing between fragmented forests or Páramo is really a focus of the social side of their conservation. Working with landowners, trying to get them to support conservation of bears on their land by supporting them.

To protect their livestock. And this converts to bears like to eat corn too. They like to eat fruit. And so, they’ll go into orchards or fields. And they can really or sugarcane, they can really do damage in a short, just a day’s worth of time. They can really damage someone’s crop. So, that is a big aspect of conservation here in Ecuador.

Just to give you a little bit of context for this species and for the challenges that we have here in this small country, but this incredibly diverse and human influence country.

Brian: Complex and challenging to be sure. You mentioned earlier, Becky that now that you have some experience with the technology under your belt you got a lot of great ideas percolating in your head.

Can you share any of them?

Rebecca Zug: Yeah. So, one, one great opportunity here is not only within Ecuador, but to start a larger collaboration across A& M Baird distribution. So, two weeks ago, I was in Canada. At the International Bear Association’s conference, they have it every two years and bear biologists from all over the world come and we share our work, and we get to, to meet each other.

Many people we’ve only talked to over Zoom are emails and we actually get to sit down together. And so, during one of these meetings with all the Andy and bear folks, we were able to talk about this project and Russ was able to share his experience working with Bear ID and the paper they published together.

And then I was able to. Start to talk to the community and let them know that to get this up to get this tool running for us. We’re going to need to collect some of these data sets across bear range. And so, getting researchers involved with each other in Bolivia and Colombia, Venezuela, Ecuador, Peru, across Andean bear range, and look at this range wide distribution of bears is one, one part of this that I’m really quite excited about Another part and research questions are it’s gonna help us over the long term to understand bear movements survival rate, females, right?

Something that Russ and I did together is that we looked at captive bears and whether or not those patterns on their faces change over time and we found that they don’t change. So, when we see a cub that’s just come out of the den, we’re able to say, okay, this is the mother, this is the sibling, and those bears, we can follow them for the rest of their lives. Because that pattern on their face doesn’t change. And so, it’s going to be able to help us with this individual ID, track these bears over long periods of time. And it’s going to help us do it a lot more efficiently. Again, right now we’re doing this all by hand and there’s always the factor of human error and the certain point it’s hard to separate individuals.

This software this this project is really going to help us make that more, that process more efficient and answer some of these bigger questions. But what happens to the bears when they disappear from one area? Do they move to the next? Do they come back over time? Because it’s easy to forget an individual you saw him, in 2015, and then maybe a different group is studying in the area, and they show up in 2020, right?

And we’ll have this nice database, or at least that’s part of my goal, is to do these bigger projects.

Brian: Ed, back to you. You obviously now have two big projects under your belt here. Becky talked earlier about the potential to expand the technology to other species. It seems somebody maybe like me Oh, this, that’s a no brainer.

You just train it on a zillion photos of Pumas and you go to town. I can’t imagine it’s going to be that easy. Talk a little bit about that and how you would like to see this work. Proliferated in other areas.

Ed Miller: The process is pretty similar as you described. It is throw a bunch of images at these models and they start to learn.

The challenge right now is the kinds of models that we’re using depend on data, which has been labeled by humans. And by that, Becky or somebody like Becky who is familiar with these populations has gone through these A B comparisons that she talked through before and. Said this is bear a this is bear B and has pre labeled them all into buckets and then we can use that as training data for the machine learning.

We call this supervised machine learning. And then this model learns those ideas fairly well, and then we can keep adding new ones as we continue. Unlike the Indian bears, pumas have even don’t really have any distinguishable markings more like the brown bears that we have, except the brown bears at least are pretty highly visible when they’re out fishing and people can watch them for long lengths of time and really get to learn the differences between these bears.

The pumas oftentimes are only being seen on these trail cameras, they’re only being spotted a few times a week or, a month or depending on the different environment and situation that it could be even longer. So, building up the ability to individually identify those pumas as a human becomes much more difficult, even than the brown bears that, that we’re working with. So, that’s the first challenge. How can we build up a data set that has been pre labeled? And then the other challenge that we can look at in parallel is there are methods for learning what we would say unsupervised or semi supervised, meaning we don’t provide the labels. We just provide enough information for.

The A. I. To start learning how to make differentiations without really spelling out what they are. And this is used a lot for projects where you’re clustering things together, where maybe you’re just showing a bunch of images and eventually the machine can learn to categorize those images into different categories.

But it’s a lot more challenging when those categories are like an individual ID of a Puma. So, we will continue to explore these kind of semi supervised methods, which maybe we can start to learn individuals by. With very few examples, or maybe we can even start adopting cross species identifications that enable us to supplement some of this data with other cats or other kinds of animals, which maybe have other ways of identifying them, like leopard spots or tiger stripes.

Maybe we can use those to start building data sets that we can blend with Puma data sets for faces. To build the technology. So, a lot, lots of things to explore, but definitely a lot of challenges I’ve had to continue to drive this into more and more spaces. Hey, life’s nothing

Brian: without some challenges to overcome, right?

Absolutely. Becky, in the academic context, as your colleagues have seen and students have seen the impact that digital technology. in harsh conditions, but just digital technology generally can have on research work. Is that starting to create bubbles of inspiration in other areas of research?

Not necessarily in, in in environmental areas, but other areas.

Rebecca Zug: It’s, and I’m suddenly do this a little bit because it all happened really quickly with Ed coming down this summer and getting this server and then starting to think about who else could benefit from this technology.

And then all these people started emailing me and asking to be taught how to use these programs that Ed had introduced to some of the researchers here. Because I think we’re all we’re all wanting to get away from spending our weekends identifying camera trap pictures, put things in safe places where we can get systems quickly and share these big data sets quickly.

So, I’m not sure if that’s answering your question too much, but I’ve seen a lot of inspiration across the biology department in lots of interests. in this kind of digital technology and getting away from the slogging through the pictures so, that we can get onto the questions. So, I work with the cloud forest research station here as well.

We’re doing a camera trap project there. It’s a new station and we’re just trying to get a baseline data on the species that are there. And then also our rainforest station here in the Amazon has years’ worth of camera trap data and they’re about to launch into a big project with jaguars, and I believe ocelots or Pumas right now and so, I invited them to participate.

In this project and to use the server and for all of us to work together to get on to focusing more on our research questions. And so, as these landscapes in Ecuador changes, the habitat changes as we see more influence and depending where you are, you have these different threats, the cloud forces mining and deforestation, like the Amazon has oil extraction and hunting and human encroachment in that area.

This kind of interest in the server in the project in general is going to help us all get to those questions a bit faster. So, that’s the inspiration that I’ve seen people trying to add this technology into their projects to help make their data processing and collection a bit more efficient to launch these bigger projects.

Yeah, that’s interesting.

Brian: We’ve done stories in the past about a company that has, I guess you’d call them audio traps. In the rainforests that will trigger for the sound of chainsaws so, that you can alert safeties to illegal logging. And I think going forward, I’m simply going to refer to Ed now as the Johnny Appleseed.

of their identification in North and South America. Anyway, the work that I think the work that you both are doing, Melanie Clapton as well, the entire community that is coalescing around this is enormously valuable and we wish you the best going forward. Thanks so, much for spending so, much time with us here.

Rebecca Zug: Thanks for having us. It’s been great to talk a bit about it.