Nathan (00:16) Today's podcast we will be talking about a study that was funded by the Department of Homeland Security Science and Technology Directorate Detection Canine Program. This podcast presents the positions of the researchers and not necessarily that of the Department of Homeland Security. Nathan Hall (00:30) Welcome everyone to the JSWK9 podcast. Today we have an array of special guests with us today. We have Sarah Kane from Texas Tech University, Bart Rogers from Auburn University, Jennifer Davis Miller from Auburn University, and Lucia Lazarowski from Auburn University. Will you guys in that order give us a brief introduction about yourselves? Yeah. Hi, I am Sarah. I am a PhD student at Texas Tech University studying specifically canine generalization. And Bart. Bart Rogers (01:03) So at Auburn, I oversee the selection and development and evaluation of dogs when they're born up to the point that they're training on their first odors, explosive odors, and foundationally preparing them to go to the field to work. Nathan Hall (01:27) Thank you, Jennifer. Jennifer Davis-Miller (01:29) Hi, I'm Jennifer Davis Miller and I'm a canine instructor for canine performance sciences at Auburn University. Nathan Hall (01:38) Lucia Lucia Lazarowski (01:40) Hello, I'm Lucia. I am a research scientist at Auburn and I am a co -investigator on this project. Nathan Hall (01:49) Awesome, well thanks guys. It's fun to actually kind of all get together in one little space here and talk about this project. So what are we actually talking about? We recently wrapped up about a year and a half long study that ended up taking us think all together where we took a whole bunch of dogs and asked a simple question of, you know, what are the best ways to imprint dogs or do that initial odor training to a variety of different kinds of odors? Do you, if you have say six or seven or eight in our case, different odors that you want to train the dogs to, do you mix some together and present the dogs as a single mixed odor? Do you train all the dogs on one odor, then move on to the next odor? Or do you train them where you sort of throw all the different odors at them at one time and let them learn them not in a mixture, but that each trial or each sort of set of odors that you're training up gets varied essentially from day to day. Does one work faster than another? Is one a bit more efficient than the other? Does one lead to better generalization? In other words, are the dogs better able to generalize to different variants? So, say for example, you train to one explosive, does that lead to better generalization to variants of that explosive than the other? And that was basically the questions that we had in this project. Lucia, you were sort running the project as sort of the co -director on the Auburn side of this. Can you give us a little bit of why did we do this study and why was it Lucia Lazarowski (03:15) Sure, so detection dogs have to learn a pretty long list of odors before they can become certified and go work in the field and that training can take a long time. Time is money. So a lot of explosives are part of the same chemical class as others and share characteristics and so if we can train dogs on just a few examples from a class and then they can generalize to others from that class without explicit training that would save us a lot of time. And also, it's highly likely that the dogs are encountering targets in the field that very different than the ones that they saw in training. So they can smell different depending on the manufacturer or the specific ingredients or other factors. And unfortunately dogs tend to be very specific and they don't always generalize. And so if we can find ways to help facilitate that, that would be very important for making training more efficient and also making them better performers in the field. Nathan Hall (04:09) Yeah, that's great. I mean, you guys have done research previously, and I think we even had it in some of our podcasts before about generalization failures with ammonium nitrate varieties and what were some of the others that you did on that as well. Lucia Lazarowski (04:22) Yeah, so we've looked at dogs generalizing from Ammonium nitrate to different varieties of it. We've looked at potassium chlorate and different sort of ingredients that are used in homemade explosives and how that can change what the odor smells like. some odors that are mixed with something else to make it an explosive. Depending on who's making it and where they are and what ingredients they have available to them, that can change the profile of the explosive. And so we've looked at if dogs are trained on one specific example or variant, then how do they respond to when it's changed a little bit or a lot and kind of the gradient of what that looks like. For the most part, we find that unless you do specific things in training to give them experience, they're very specific to what they know. which is not surprising because dogs have really good noses and it's probably because they, any slight change, they're going to pick up on and detect. But that can be challenging when you can't train them to everything that they're ever going to encounter because you never know what they're going to encounter. so finding ways to help bridge that is really important. Nathan Hall (05:29) Yeah, exactly. So definitely getting that good generalization can potentially have not just even on training efficiency, but a lot of sort of consequential impacts. So this project, how do we actually do it? Sarah, you were sort of one of the people doing the day -to -day testing. What did it actually look like? How did we, how many dogs did we have? How did it break down? So we began with two different groups of dogs. We had dogs from Auburn's Canine Science program, and then we had dogs from Texas Tech's Shelter to Adopt program. Both sets of dogs were completely green on the odors that they were being tested on in this experiment, but they did have varying odor experiences outside of the study. So the Texas group was, for our knowledge, totally green to any odor detection work. And some of Auburn's dogs had some previous experience on odor detection tasks. Again, not with these odors. And so we worked these dogs in different cohorts. And we ran four cohorts through these experiments. And we did one experiment where we looked at generalization to a series of odors that were within the same class, so smokeless powders. We looked at dog's ability to learn and then also generalize to novel types of smokeless powders within the same class. And then we also looked at between classes. So we had explosives from ammonium nitrate based, PETN based plastics, as well as TNT based explosives. And so then we looked at the acquisition rates, so the how fast a dog learned an odor and also their ability after these different trainings to generalize to novel targets. Because again, generalization is a critical skill for a lot of these dogs in the field. So probably the first question people want to know is, was one faster than another? Yeah. So first looking at the between classes of explosive experiment, we did find that The intermixed group, so the group that learned odors, they would learn four odors at a time, but they would not mix the odor head spaces. Each odor was presented discreetly and trained discreetly, but they would learn four within one training period. Those dogs were slightly, although not significantly faster to learn the targets that we were asking them to learn compared to a mixture group. And in the mixture group, those dogs were trained to the headspace or the mixed odor head space of four different explosives. And they were slightly faster, but not significantly so, than a group of sequential dogs, which were trained in the method of learning one target at time. And then once you acquired that target, moving on to the next, which is how a lot of dogs in the field are trained today to detect different targets is one at a And so we found that that intermixed method did lead to slightly faster acquisition rates, but not significantly so. Nathan Hall (8:21) So then for the mixture group though, you had four different odors being mixed together and you trained them to that. And then you tested each of those, you sort of pulled it out and tested whether they responded to each of those individual odors. What were those four odors and what were the dog's responses to that? Sarah Kane (8:39) So we had one variant from each class that I mentioned earlier. So that was plastic explosive, ammonium nitrate, PETN -based, and then TNT -based explosives. And so we had a mix of the headspaces of four variants of each of those classes of explosives. And we found largely that dogs that were trained to that odor headspace mixture largely just responded to our plastic -based explosives. for this, it was C4. We saw pretty strong generalization to that from just learning the mixture, but we saw relatively low rates of generalization to PETN -based ammonium nitrate and TNT -based explosives within that group. So how many dogs spontaneously hit to the TNT? Just out curiosity in that mixture. No dogs spontaneously hit to the TNT to meet criterion on TNT, meaning all dogs had to be explicitly trained to search for TNT. They couldn't go from the mixture to automatically hitting on TNT. What about smokeless powders? Because in the smokeless powders you have four different varieties, right? Of single bases and double bases of smokeless powders mixed together. Did that, how many did they generalize from that mixture? So yes, so to go back a second for our within classes of explosives, we looked at smokeless powders and we saw that acquisition rates actually didn't change between any of the groups. All dogs, regardless of their training in an intermixed fashion, which is again, presenting odors discreetly, mixture or sequential, those dogs all learned those initial targets roughly in the same amount of time, not significantly differently. The dogs in the mixture group, they were able to generalize to three of the four components of their mixture target, which did facilitate strong... acquisition so they could learn three out of the four relatively quickly. However, that didn't actually lead to faster training overall the whole period. It just led to a jump initially. And then they still had to learn the odor that they failed to generalize to initially. Nathan Hall(10:39) So overall, a zero -sum game basically, right, in terms of learning time, has kind of all worked out the same. Yes. And what about generalization? Did one group tend to generalize better than another? Sarah Kane (10:50) Yeah, this is actually where we started to see some results for the different training paradigms. We did see that the dogs that were trained in an intermixed fashion, and this depended a bit on the odor class that we were looking at. So for our TNT -based explosives, our intermixed dogs actually did have a 72 % hit rate on TNT variants that they had never seen before in training. And that compares to a 25 % and a 28 % between the mixture and the sequential groups. So we did see an operationally relevant hit rate advance for our intermixed dogs compared to our mixture and our sequential dogs just for that class of explosives. We did see significant results for generalization or higher hit rates for plastics and PETN based as well. However, these results were less operationally relevant. So we saw that intermixed dogs had a 19 % hit rate for some plastic -based explosives compared to a 3 and 6 % for the sequential and mixture. So operationally, that may not be useful for handlers in the field. But for TNT -based explosives, we did see that generalization rates were better after intermixed training. Nathan Hall (12:06) Did you see more of a change of behavior for the dogs that were intermixed versus the others? Sarah Kane (12:12) So if we looked at sniff rate, so that's the amount of time a dog spent sniffing the odor, we did see that for, there were changes in behavior for the intermixed group compared to the other two the mixture and the sequential groups as well. Okay. So even though they may not have had a high, as high of a hit rate, was still at least some change that maybe might make it easier for training, but it suggests that you can't just intermix and then get spontaneous generalization to everything. You still have to train to a lot of those variants to actually get what you want. So it's something that helps, but is not the, I don't know, the secret sauce, I guess, if you will, the thing that just kind of. It's not a magic bullet. Exactly. Cool. Is there anything else you had in there? Yeah. So this actually goes to Lucia's point earlier that we also were able to look at if different volatiles in the headspace of explosives helped explain some of these generalization results we are seeing for the dogs. And for both experiments, that within class and that between class, interestingly what we found is that odors that were unique to certain odors or certain explosives that we were using actually made dogs less likely to generalize to those odors. So it wasn't about commonality, but it was about what made things distinct that could drive that generalization. So for example, if you had a contaminant in one of your odors, the dogs are less likely to generalize to that compared to even if it has a lot of the commonalities of other types of explosives that they may have been trained to, that contaminant may actually drive their lack of generalization. Nathan Hall (13:48) That's interesting. So I always like to equate things all back to foods to bring sense and taste back to food. And pasta sauce is always a good one, because everyone's got their own kind of pasta sauce with different ingredients. But if you take your pasta sauce and add cinnamon to it, it's no longer a pasta sauce to me. Even if they both have tomato in it, if you have cinnamon in one of them and add it in, it's still a different, it's not the same thing. I wouldn't classify it as pasta sauce anymore or something different. Lucia Lazarowski (14:11) It's chili now. Nathan Hall (14:13) Exactly, Yeah, I know it's completely different. Exactly. That's actually really good because ground beef and everything all could be the same difference between chili and pasta sauce. Yeah, that's a really good point. Lucia All right. Awesome. We're also, you know, as part of this really lucky that we had, you know, a researcher's perspective, but also some, you know, highly skilled expertise from actually working with operational dogs. and experienced handlers working with us from the Auburn Canine Performance Science Center side. So the Auburn trainers took this intermixed method and have been working on sort of trying it outside of the lab or taking it out of the box, if you will. So what does this actually look like in terms of putting it into practice using sort of it as a real world imprinting method? Jennifer Davis-Miller (14:57) Well, when we put this into practice, we chose four target odors as our base, similar to the experiment. We imprinted each of those odors individually within a training session of 10 trials, assisting the dog's trained final response for the first few trials, and then allowing the dog to work a little more independently, still assisting if needed. With the intermixed method, you'll start layering odors before the dog reaches proficiency on any one target. So during the initial imprinting, the handler should be able to see a noticeable change of behavior or an independent trained final response from the dog before moving on to the next odor. Once it's clear that the dog is recognizing odor, you'll layer in the next odor. And once the second odor is imprinted, you can start using both of those odors as targets within a training session, alternating between the two. Likewise, after we imprinted the third odor, we alternated between all three of the targets and so on until all four odors had been imprinted and alternated in training. Nathan Hall (16:02) So how did it work out? Did you like it? Jennifer Davis-Miller (16:05) It was great. Bart Rogers (16:05) Yeah. Nathan Hall (16:06) Yeah? How did you think that that compares to what you might have done previously or in other methods? Bart Rogers (16:12) Well, and traditionally, and how I've done it for years up until this point, and some other previous things that did it unintentionally I didn't know I was doing intermix training when I did it. But roughly been involved in somewhere around 500, 600 dogs imprinted on multiple odors. And typically what we do is you reach full proficiency, right? So we're working, we're starting with an odor. Nathan Hall (16:39) Mm -hmm. Bart Rogers (16:42) first odor, second odor, you would work the dog until like, okay, let's run criteria on this dog. He has to do 20 unassisted trials, right? No falses, no misses, change of behavior, full final response, right? And then what you would do would go because of spending so much time in sequentially training it, it would be the more sequential method, is then you, we've got to worry about transfer of context. Nathan Hall (17:12) Mm -hmm. Bart Rogers (17:12) Once the dog meets criteria, reaches proficiency on the odor, then we would take that typically to an intermediate step, some other type of box. It's not a box lineup of imprinting and work there and then move from that and then to some elementary level operational search set scenario. And then. we would probably take it two different operational search scenarios, three or four different types of operational venues, and then come back and layer your next odor, right? So, you take that in consideration of by the time I'm getting on, some places may have required 11, 12 odors, some eight, typically around eight, you've repeated that process. It may get quicker, but by the time you're on that eighth odor, Nathan Hall (17:44) Mm -hmm. Bart Rogers (18:04) you may be having to cycle through, not only is it hitting this odor, now I have to bring out a scenario where I layer in the odors that I already know in a search right? That I'd learned previously. So the difference would be that we're training four odors, right? We're not spending all the time in meeting this level of proficiency before we shift to the other one. And instead of a matter of say two to three weeks, two weeks to layer four odors, we now have four odors And what would you say, Jennifer, we were easily with a quality dog, two days, two to three days? Jennifer Davis-Miller (18:38) yeah, yeah, absolutely. And that was only running two short sessions a day with each dog. Nathan Hall (18:41) So. Bart Rogers (18:45) Yeah. So being able to, and that's with, you know, testing, whether they, you know, running a baseline test to actually know that they know the odors once we finished the four odors, right? So actually checking that at the end. So not only was it, you know, how much time and efficiency you got from that logistical, you know, issues with sequential, the way I've always done it, you know, we're down two or three days. I've got four odors. Nathan Hall (19:12) This sounds like one of those really weird situations where things work better outside of the lab than in the lab. Where we found something that seems to work and then usually when you bring it out to the field and test it, it all falls apart and disintegrates. But it sounds like it's actually getting better with time rather than worse. Jennifer Davis-Miller (19:17) Hahaha Bart Rogers (19:28) Well, I I think that's always the fight between the trainer and the scientist is, you know, we have we have to control so many variables to say for certain what the data means and, the integrity of the data. Jennifer Davis-Miller (19:32) Yes. Bart Rogers (19:38) When we loosen this up, now that we know from the data into a practical aspect, in a practical type scenario, we're able to move a lot faster. I think outside the constraints of the research, I think you'll probably see stronger results in it. Well, we're not done yet, but I know you would. Nathan Hall (19:55) That's great to hear because you it's a you always end up in these situations where it's like, yeah, it worked in the lab. But then when you pull it out, it doesn't work at all. it seems like now that that loosening is actually just allowing you to make it even stronger when we don't have to, you know, isolate and control every little tiny variable that we were before. So that's great to hear. So I was thinking, as you were describing that Bart conceptually, is it almost seemed Bart Rogers (20:16) Yeah. Nathan Hall (20:21) I know, is it like keeping the dog on their toes or on their paws, I guess, if you will, in terms of always mixing that up so that they kind of always got to be on Bart Rogers (20:34) Yeah. I mean, I think we'll, you know, Jennifer we'll discuss, you know, some things that we did, we did, you know, taking consideration with this, but yeah, from my perspective is it's the mindset of, know, you have a high quality dog, you know, our dogs are purpose bred. They're, you know, we breed them to work. We develop them to work and a dog's nature, you know, coming from, you know, their ancestors is to be efficient, right? And to be efficient is, you know, is lock into something that is the easiest to get you rewarded. If you make mistakes, then I'm going to find something that I can guarantee 100 % of the time I'm right. So sequentially training, like we traditionally do, reaching the 100 % proficiency, that is more reinforcement and more trials to the point that that dog is so specific, right? And then we're always worried about contamination. Dogs picking up on things that are not the explosive odor. So we're using pristine samples, correct? Because we want to present pure odor. And with that much reinforcement history with it, we you know, I have seen this and experienced it with with really high quality dogs and litters that they can even be so specific to say smokeless powder that they'll only find one smokeless powder that is from the certain batch and lot number from the factory. You know, that I've seen Nathan Hall (21:37) huh. Yeah, yeah. We even saw that in some of the experiment. It even just changing out the vial a little bit would then cause some generalization of a decrement. This is used versus unused smokeless powder. Jennifer Davis-Miller (22:04) Yes. Bart Rogers (22:07) Exactly. So, know, keeping that mindset, not enough reinforcement, reinforcement to generally know the odor, right? Generally know the odor. And then that odor can change as we layer things, right? So not only are they not too specific, but they are always thinking there could be another type of odor. And probably, know, a fringe benefit of this or, you know, unexpected benefit is a search technique, right? Jennifer Davis-Miller (22:10) Yes. Bart Rogers (22:39) So not only, so if I start with certain odors and this dog's foundation is easier odors, because I'm not gonna start with the hardest odors if I'm sequentially training green dog, then their whole foundation and hundreds of reinforcement of learning odor is working in a way that is not systematically and methodically interrogate in a way that they can detect harder odors. Now when I get to a harder odors, I'm not teaching a dog an odor now, I'm fighting the procedure and the way it's learned to operate and the task that's working against it. So there's a lot of benefits from that. That's not just maybe generalization. Nathan Hall (23:11) Yeah, are we? We saw some of that almost in the lab too, right? When you get some of the smokeless powders, dog can just kind of pick its head up and it's there. That's not a problem, but that's not going to work for TNT, man. You need a different search technique for that to work. Bart Rogers (23:24) yeah, yeah, yeah. I don't know. Yeah, I mean you go in from smokeless power. Say I did this a long time ago when I didn't know what I was doing. Start with C4 and then my next one we're going to TNT. It took a long time for these dogs to get to where I could get them to work to detect it. Nathan Hall (23:46) Yep. So have you taken this intermix out from boxes to like searches yet? Bart Rogers (23:55) yeah, yes. I mean, Jennifer can explain, the, setup we did. Cause that, that was one of our things was we also know, you generalization is an issue, but also we're always fighting with sequential training is transfer of context. Nathan Hall (24:10) -hmm. Bart Rogers (24:11) We're going from boxes, our boxes are just to learn the odor so we can have them located in an operational environment. And as I said before, traditionally you would, you would do an intermediate step, right? And depending on the quality of the dog litter, you may have another intermediate step and then multiple venues before you lay another odor. So Jennifer, she, she can explain cause you know, she designed, how we, how we mocked up that transfer of context where we still did have some control at. to clearly see the results. Jennifer Davis-Miller (24:43) Well, when we did this experiment outside of the generalization boxes, we did use an odor lineup for our traditional imprinting. But then when we took it outside, our context changed. We had a couple of different setups because we were using some lower volatility odors. So it had to Be careful with things like that, also distractors. So we tried a few different setups before we got to one that we liked, but we used some cinder blocks. We did kind of a couple different versions of some odor walls that we liked. So we changed it up kind of with the mix of the dogs that we used. Nathan Hall (25:20) and it seemed to work across those different contexts pretty well. Jennifer Davis-Miller (25:23) It did. mean, there was a learning curve at first with the very first setup we did with this odor wall specifically. The odor was kind of traveling up the wall through seams that we weren't anticipating. So the dogs weren't alerting right where the odor hole was. They were kind of alerting on the sides and on the seams. So we changed things up a little bit so it wouldn't be quite so frustrating for them, but it was good. Bart Rogers (25:49) Yeah, so in short, we got recognition of odor. That wasn't an issue. So we saw our transfer of context. So when we redesigned it, we redesigned it that there that in our line up or wall that. Nathan Hall (25:53) Mm -hmm. Bart Rogers (26:06) when the dog hit odor and was showing us where source was. Source was where actual source was and not drifting down to now I've got a dog responding and I know that's change of behavior on odor and responding to odor, but there happens to be a distractor or a novel thing because we need to rotate that in to be sure in that general vicinity. So we are through through the development of it. We we had the dogs work a lineup that they Nathan Hall (26:22) Mm -hmm. Bart Rogers (26:34) never seen before. Like they've only seen boxes. Never worked cinder blocks. Never worked paint cans holding the odor, right? In an area they had never worked before outside of the imprinting area and these dogs had only been in one room doing imprinting. So off the truck, completely different area, completely different setup. And I would say, Jennifer, you want to explain what odors we used and the variants of them. I think you already said the Nathan Hall (26:43) Mm -hmm. Bart Rogers (27:03) the odors, but what type of variants we used to probe. Not only, we got the transfer of context, that they hit the odors that they had learned, right? But then we wanted to look at generalization, now we knew the transfer of context. Nathan Hall (27:13) huh. Jennifer Davis-Miller (27:17) Right, so our base odors, we had a single -based smokeless powder, double -based smokeless powder, AN and also TNT. So the variants that we used were, you know, one from each of those classes. And then what we also did was take a variant, I believe we used smokeless powder and we changed the packaging. Nathan Hall (27:30) huh. Jennifer Davis-Miller (27:38) So all of our odors during imprinting were packaged in a cotton bag. And then when we exposed them to a variant of that odor in their contextual change session, we packaged them differently. So we had it in an anti -static container. For one, I believe we had one mixed with a piece of metal pipe. So just kind of giving them the same odor, but taking, removing the packaging that they were imprinted with and putting it with something else and seeing if they would respond. And we had a really good success rate with Bart Rogers (28:09) Yeah. And for, you know, for say one odor, in particular the TNT, as we found in the gen boxes, when we were doing the study that we had some contamination, right? We had some plasticizer contamination from... Nathan Hall (28:09) That's fantastic. Bart Rogers (28:26) handling use of it in the past that would probably be present and realistically in every one of the TNT they're using. And we presented that TNT which we knew was contaminated. And as you said before, you had saw in the study that, know, that that contaminant may prevent the generalization. Well, this explosive was, the TNT was housed in a different barrier material. Like you said, they had only seen this barrier, all of our explosives in one type of barrier material. On top of that, it was contaminated. we saw a correct me if I'm wrong on Jennifer, no dog questioned that. We saw no dog questioned anything, to my knowledge, except for the cast TNT material. Jennifer Davis-Miller (29:00) Correct. Yes, and I believe we had one single -based smokeless powder that a couple of the dogs, maybe it was just one dog, that bypassed, but the rest of them showed at least a change of behavior, if not an independent trained final response. Yeah, the dogs did really well overall with the variants. Nathan Hall (29:24) Turn it Bart Rogers (29:27) And the cast TNT would be expected. I mean, think everybody's experience and maybe research would show that going from flake to a cast TNT material is usually explicitly trained and treated as a completely different odor. But we did have dogs hit. We did have one dog hit every odor. Now we could say that we had novel distractors. We controlled for that when we did the shift. She also did hit, I believe, one novel loud chemical odor. So Rustoleum. And we broke the cast material down too. We even had a version of it that was ground instead of just the pure cast material as you would see it naturally. Jennifer Davis-Miller (30:00) Yeah, it was either the Sharpie or the Rustoleum, one of the two. Bart Rogers (30:12) So that shocked me and even with the cast material I believe we did see a lot of the dogs or some of the dogs have a change to it. And if you had asked me through all of my experience of going from flake TNT to cast TNT you wouldn't even get a change in their first experience. Nathan Hall (30:29) Yeah, I think our control groups in the Gen boxes were like 10 % response rate to the... I think so. Yeah, yeah, so definitely a big difference there. That's cool. If you're going to apply this, you know, for other people listening, we're going to apply. Are there any considerations that, you know, did flag up that you would say this is what you need to think about if you're going to try and apply this in your own situation? Jennifer Davis-Miller (30:56) I think it would definitely be beneficial to do some task training, some foundational task training with the dog so they understand how to work a box lineup. Definitely that. Training with relevant distractors would definitely be another factor to keep in mind. Items used in the packaging and storing of your training aids, gloves, novel odors the dog might encounter in an operational environment, those are all going to Nathan Hall (31:05) Mm -hmm. Jennifer Davis-Miller (31:21) good distractors to use in your training early on. Bart Rogers (31:24) Yeah Nathan Hall (31:24) That's a fair point because if you've got something that's going to help you with generalization, if you're not careful in that imprinting to get your right distractors in there, like if you never use handling material or you never use any of the gloves as distractors, you might actually be going too far in that direction. So it might be quite important to consider having those right distractors from the get go. Bart Rogers (31:49) Yep. And one thing that... On top of you know having the component distractors and contamination how it's handled very much barrier material as we saw firsthand Nathan we went you know back and forth on this and figuring it out which made sense is is thinking about the strengths of the target odors and and how relevant are the Distractors if we're calling distractors when it comes to signal strength, right? So if I you were working with c4, you know, your relevant distractors are going to be things that are more loud, you Nathan Hall (32:14) Mm -hmm. Bart Rogers (32:22) chemical plastic things, things of that nature. Now when you get down to TNT and AN. you need to have your lineup and basically, Jennifer developed the protocol into how the lineup was to cover all those for, say, those four. So we had a blank box and we had things that, know, maybe the burial material was nothing, which is yeah, it's a barrier component, but it's a subtle different odor and then loud odors. So when you get to the AN and TNT, your most likely and most relevant distractor would be the box with a subtle odor to it, maybe the barrier material. And then the most, Nathan Hall (32:55) Mm -hmm. Bart Rogers (32:57) relevant would be the blank right The signal strength of the distractors is one thing, on top of them being components and contamination of barrier materials and glove, the strength of which the odors you're training should be taken into consideration. So if you have loud odors, you need to have relevant loud odors as distractors. And then when you get to the lower odors, lower strength odors like AN and TNT, your blank box, or your box holding something like just a barrier material like cotton bags, the subtle ones, are the most relevant. your dogs are not responding to a disparity in signal strength, they are responding to the odor. So no matter if we're doing an intermix, we need to ensure they're learning the odors and not difference in signal strength. So when we go and test this, we know they understand the concept we're trying to put into them by intermix and the odor. So we can test it. Nathan Hall (33:53) Yep. And we saw that in the research trials too. mean, some of the hardest ones were blank blank TNT or blank TNT blank or those are some blank blank blank was actually the hardest. Yeah. That's a good point. All right. So then in terms of, you know, for explosives and traditional explosives training, in your experience, what benefits do you think the intermixed training seems to have over some of the other methods? Jennifer Davis-Miller (33:58) Yes. Lucia Lazarowski (34:04) Yep. Nathan Hall (34:21) particularly the sequential that you've used previously. Lucia Lazarowski (34:24) Well. Bart Rogers (34:32) One thing you hit on from my perspective from explosives in the gen boxes, as we laugh, it's like we may see better results outside the lab. You talked about sniff rate, right? You might not have got a full on trained response, but given the sensor that we use, the laser beam and break time, we calculate that they were interested in it. They recognize some form of similarity. And that's what I call in my world a change of behavior, right? So traditionally in sequential training, I need to be able to know can they hit this type of TNT, can they hit this form of an RDX explosive for different type of C4, tag it, untag it, know, those type things. And when I get, when I place those out in the field, right, I will, when I get there, the dog should have a strong enough behavior or give a trained final response. If it does not, I need to change the behavior to assist them, right? A strong change. But sometimes what happens is in the sequential type training is that they're so specific that when you get there, you don't have a change of behavior. So now you may have to go back and train for that specific odor just like it's a different one. Okay, well that takes a lot of time. And how many variants of that odor could be? Be a lot, right? Or how many departments, how many have access to these type of variants that could be threats that their dogs need to have experience with? Nathan Hall (35:28) Yeah. Bart Rogers (35:46) only have a day out of a year or quarterly, right? So when they get there and they don't have a change, they're probably bringing that dog back to it, right? And they're going to assist it. What are they assisting it to? An environmental contamination that the could pick up from setup of the scenario, disturbance, right? Nathan Hall (36:00) Mm -hmm. Bart Rogers (36:06) So now we've got a dog that may not be responding to explosive odor. We have a dog that we are now training to detect disturbances in a training scenario that will not be there in an operational setting. But if I come from this type of training and I get a change of behavior based on the way that we did it with these variants. I can't assist the dog and guarantee that that dog is bridging the gap and now he goes, now this scent picture can be this much different, but I may not have had that ability to do it without this style of training. And if I didn't, I have caused more problems if I assist the dog and know that it's probably not odor, right? Nathan Hall (36:41) Mm -hmm. Bart Rogers (36:47) Even from a handler cue perspective, like we want these dogs to be independent. There's no change of behavior. I don't have access to this. I've got to get experience with it in this ape placement. I'm bringing it back, cueing the dog, telling, hey, check here, check here, and just listen to me and sit. OK, now all this training of being independent and operationally relevant, I'm backtracking on. And I'm probably not teaching what I think I'm teaching to the dog and making the dog less operationally relevant. Nathan Hall (37:15) Yeah, I think that's a really good point that maybe we hadn't really thought through from the lab side completely is because, you know, even though we're getting that increased change of behavior and we're thinking, well, but it's not giving us that full four second nose hold. But if you're thinking of it from a training perspective, that's got to be massively different what it looks like in terms of difficulty of training when it comes to a dog that's going to turn their head, hold their nose for two seconds or two and a half seconds. versus a dog that walks it, like this has nothing related to what they're doing whatsoever, that one dog you can shape in a matter of minutes and the other dog, you're back to square one, Bart Rogers (37:55) Yes, and from an operational standpoint, not just training for an operational standpoint. If you're searching with a dog, and most people will tell you this, trained operationally, yeah, trained final response is great. But if I've waited for that to happen, I've messed up, right? Most of the time, I already know where the target is. I've read my dog's behavior, right? So what you would see with it, maybe the odor is different, but the dog recognize it based on your imprinting technique. And you see some generalization, but the dog gets to source, works odor gets to source and say, there's some things different here. There may be some things missing. There may be some things added on that's different in this variant. And he may not give a final response. He may show a super strong change in behavior and he goes, yeah, that's not exactly right and leaves. Nathan Hall (38:32) uh-huh. Bart Rogers (38:39) But what would happen in operationally is I'd see that super strong change of behavior and say that is my dog's behavior and odor, right? And it's right there. We're already out of there. That's somebody else's problem at that point, you know? But that's a fine. That's what real order fines look like. We train for the worst case scenario so that they give a passive operant response to be safe in the presence of a explosive. But in reality, I would never want to get to that point. Nathan Hall (38:51) Yeah, exactly. Yeah. That's good point. That definitely highlights the importance of bringing that trainer perspective in on some of these paradigms as well. So we talked a lot about explosives. Do think there's applications of this outside of explosives? Jennifer Davis-Miller (39:23) absolutely. For me personally, outside of the work that I do at Auburn, I train human remains detection dogs. And with this type of detection training, the dog's ability to generalize is very important. So I use an intermixed method with my own working dogs when I do their initial imprinting and have had a great deal of success. I can imprint a sample of tissue, for example, and then continue to layer target odors with different levels of decomposition. So I'm able to imprint a broad spectrum of odors in a short period of time and then can use those odors within the same training area. And the dog understands that there are multiple targets that they can locate. Bart Rogers (40:09) And I'll add to that, in my experience with, you know, in some of the projects that I get to do here in conservation work, you know, in novel pilot studies with, say, for example, pythons. Like, we got asked to do that years ago. We have no idea whether a dog can smell a python. What odor would they be detecting? And what I wanted to not find... native snakes in the areas in south part of the Everglades, right? Some of them are venomous and there's lots of other non -venomous species. So, but I needed them to generalize because we wanted to find hatchlings, neonate hatchlings may have just hatched, never ate before, maybe that's a factor. We wanted to find juveniles of sex, male, female. We wanted to find mature individuals of male and female. And then there's a seasonality effect of when they go into breeding season where the males track down the females and they have mating balls. So something's changing there scent -wise for Bart Rogers (41:03) to find them and how do we do that? And then when you're working with a highly regulated invasive species or say maybe you have endangered species, training aids are at a premium, right? Especially when you need a training aid outside the area that they don't have them so they don't want you to take them there. So what we did was broke down age, know, juvenile, mature breeding age, male, female, and then we broke it down Nathan Hall (41:14) Mm -hmm. Bart Rogers (41:28) the area on that individual. We collected samples odors, know, dorsal, ventral, whole body, head, and even snakes that were in blue phase, so they're about to shed, things of that nature. And I've got six snakes, and I've got to use these snakes to train the dog so that I can transition to the field, right? Nathan Hall (41:47) huh. Bart Rogers (41:49) And I've got to use that to train to get there to South Florida. And if I train too much with those individuals, I'm least likely to bridge that gap to the odor of the operationally relevant target, right? The more I train. So how do I slow play it? And how do I ensure I get generalization but not overgeneralization? Bart Rogers(42:11) I did intermix then, I didn't know that's what it was. But it was just breaking it down. How would we get generalization, not generalization? And it worked. We found everything other than a hatchling. But we did test hatchling in plot studies and they did hit hatchlings. I'm pretty convinced that I never came across hatchlings. So yes, there is a benefit from being able to capture not just males, not just individuals within a species but to generalize but to generalize in an area that have similar species of snakes or other species but not your non -target species. So the generalization range to maximize the dog's potential to find the targets or any individuals in the target but not to overgeneralize and get false responses. Nathan Hall (43:00) That's cool. I mean, same problem, different target, know, other complex ways. Thank you guys very much. mean, I think that covers a lot of that project. But you have fun, that's the most important thing. Bart Rogers (43:14) yeah. yeah. Yep. Jennifer Davis-Miller (43:15) It's been a lot of fun. Nathan Hall (43:16) Well good, well thank you guys so much for joining us. Thank you for giving us a rundown of this project and more details to come out in print and publication as it comes through. But thank you for your time and helping explain this and giving us the updates on how this is looking, transitioning to operational activity. And glad to hear it's only getting better and not worse. Jennifer Davis-Miller (43:38) Hahaha Nathan Hall (43:41) I feel like that's not always the outcome of research, so that's great to hear. Jennifer Davis-Miller (43:45) Thank you, Nathan. Lucia Lazarowski (43:46) Thank you. Nathan Hall (43:46) Thank you guys.