Puppy Forum and Dog Forums banner

1 - 12 of 12 Posts

·
Registered
Joined
·
192 Posts
Discussion Starter #1 (Edited)
Sorry if this has been discussed but a quick search didn't yield results

here is the question. Do dogs learn through positive reinforcement quicker with more reps of medium quality treat or less reps but with very high quality food.

lets say you have two treats, both of which the dog is willing to work for but one is much higher in value. Does it take less reps for a dog to learn the skill with a very high value treat EVEN if the lower value treat is enough? and lets say Fido learns sit with 10 reps using chicken liver, and 15 reps using kibbles. (assuming the end result is the same), what about fading the treat? is it easier to fade when the skill is learned through a lower value treat than higher value treat? is it easier to fade when the skill is learned with less reps on high value treat or more reps on low value treat? and what about memory? can Fido remember the sit 3 days later if he learned it with high value treats with less reps or low value treats with more reps? (again, assuming the end result of both is the same)

in other words: is there a DIRECT trade off between quality and quantity of the primary reinforcer? and if so, how does this trade affect learning, memory and fading of the primary reinforcer

any thoughts? both anecdotal or scientific are welcome.
 

·
Registered
Joined
·
3,193 Posts
All dogs are different. I think the answer to your question will vary.

One of my dogs works extra hard for special stuff, so for her, I will get more intense effort for better stuff.

My biddable male will work for kibble, the same as for good stuff. He's super easy.

My whippet pup will ONLY work for great stuff. I can go to he## if I have regular crap. We will work through that, but that's where we're starting from.

In an effort to control feeding costs and negative nutritional impact, I blend great stuff with kibble in a container. It increases the value of the kibble.
 

·
Registered
Joined
·
192 Posts
Discussion Starter #3
I don't think the answer vary, at least not in the way you described. What does vary is what is considered to be a highly reinforcing for different dogs. Like you said, for some, kibbles may be highly reinforcing and for some other dogs, that may be cheese, or even frisbee. My question generalizes to the whether there is a trade off between reinforcement quality and repetition.

So if a dog sees cheese as uninteresting and not reinforcing while dry kibbles as highly reinforcing, then dry kibbles would be considered high quality training threat for that dog. So yes, there is variability in what kind of treats is highly reinforcing and what kind of treats is not. Its partially my fault for the language I use (calling it high quality treats), I should have been more clear in calling it "highly reinforcing stimuli" or "high value reinforcer".
 

·
Registered
Joined
·
3,193 Posts
So to see if I understand your question, let me see if I can rephrase it:

Will dogs learn more quickly with fewer reps with a high value reinforcer or more reps with a medium value reinforcer? Is that your question?

I still think there will be considerable variance as there are high rep dogs and low rep dogs. And dogs with long attention spans and short attention spans. I would think that low rep learners with short attention spans would do better with a high value reinforcer, but I am only guessing.
 

·
Registered
Joined
·
192 Posts
Discussion Starter #5 (Edited)
So to see if I understand your question, let me see if I can rephrase it:

Will dogs learn more quickly with fewer reps with a high value reinforcer or more reps with a medium value reinforcer? Is that your question?

I still think there will be considerable variance as there are high rep dogs and low rep dogs. And dogs with long attention spans and short attention spans. I would think that low rep learners with short attention spans would do better with a high value reinforcer, but I am only guessing.
yes, you are right

there is variance between dogs. but the difference is, it takes Fido 10 reps to learn a good sit with cheese(High reinforcer), and 20 reps for the same sit with kibbles(Low reinforcer). But it takes Max 20 reps to learn a good sit with the frisbee(High reinforcer) and 35 reps with tennis ball(Low reinforcer). The difference may be in the type of reinforcer, the number of reps it takes. But the general concept doesn't change. All dogs have low value and high value reinforcers. therefore, all dogs have different levels of motivation to learn based on the value of the reinforcers. so yes, dogs have different learning capabilities, attention span, etc etc. but PR learning is the same.

also to add to your last comment, I think all dogs will do relatively better with high quality treats than low quality treats in the same number reps. and it may be the case that attention span or some other factor contributes to how well a dog learns with the two different reinforcement schedules, so that would be interesting to know if anyone has any insights.

although, I feel like the relationship is one thats "basic" enough that there shouldn't be any between dog differences, but I really don't have a good grasp on it.

and how it affects not just learning, but retention and fading of the primary reinforcer is a more complicated question.
 

·
Registered
Joined
·
12,574 Posts
In Oliver's case, he learns better with more reps of a medium quality treat. He needs more encouragment. Carsten on the other hand won't care for the medium quality treat and will only work for a high value treat. So... each dog is different. That said the best case scenerio for both is lots of high value treats. :)
 

·
Registered
Joined
·
9,366 Posts
I believe that Ian Dunbar is the authority for this question, related to lure and reward. Google for details.
1. Consider the treat to be a lure and a marker of correct behavior. When you are first training, especially the very first behaviors, such as Sit, a lot of tiny treats is much better than one large treat.
2. When the dog finally "gets it," and performs on cue, then you give him a jackpot of 3 - 4 little treats or a high quality treat, as a reward more than as just a marker.
3. When you want a dog to perform a scary behavior, or approach someone/something he is a little afraid of, then you can lure him with a high value treat (There is a specific method for this), then you repeat the behavior using a lot of tiny lower quality treats (They don't have to be lower quality, but you no longer need boiled chicken or liver to get the desired behavior.)
4. Obviously, some dogs are not food-driven (can't prove that to a Lab owner :) ), and you use a high value object, like a toy, instead of food. Ian Dunbar will lure a dog to do a series of behaviors - Sit, Down, Stand - and in a few second, extract a few repetitions with only one treat.
 

·
Registered
Joined
·
348 Posts
I think you would really enjoy Dr. Pamela Reid's 'Excel-erated Learning: Explaining in Plain English How Dogs Learn and How Best to Teach Them" because it answers things like this! It's a little science-y (not really plain English), but if you're already wondering about things like this, I think it's perfect for you. It's one of the best learning theory resources I've ever had. http://www.amazon.com/Excel-Erated-Learning-Explaining-Plain-English/dp/1888047070/ref=sr_1_1?ie=UTF8&qid=1324593928&sr=8-1

Scientifically answering your first question part would be extremely difficult -- I'm having trouble designing an experiment that would discover an answer. Like trainingjunkie, I, too, see a lot of variables, though different ones.

Ideally, like you said, you'd want to teach a dog how to sit with a medium reinforcer and then teach the exact same dog how to sit again in the exact same way with a high value reinforcer and compare the times/repetitions. You obviously can't do this though because once a dog has been trained to sit he's been trained to sit -- you can never get back to the completely ignorant dog again.

You could train 5 behaviors with a medium value and 5 with a high value... but the results from that wouldn't be accurate because the behaviors are different -- "speak" and "touch" are often much easier to train than "sit" and "down" ... and some dogs would learn better with luring or with shaping, etc etc.

I don't have a lot of personal experience with treat-fading because I generally don't lure with food (I train a hand touch instead and use that for behaviors I want to "lure"). For the times that I have used food lures, I haven't seen a significant-enough-to-remember difference between high and low value treats. In general, by the third or fourth successful lure I've removed the food lure and I've never experienced a "food lure-wise dog" or one that is hesitant to work without the presence of a food lure.

A 75% relevant excerpt from the aforementioned book:

"The amount and the quality of the reward (the palatability) have direct effects on behavior: animals work harder for larger or more tasty rewards. Also, if given a choice, animals prefer several small bits of food to one big bit of food, even if they both add up to the same amount. This is because the act of engaging in eating is itself very rewarding.

Conducting preference tests with dogs is a difficult proposition. If you were to sit down and show your dog a variety of different food treats and then allow him to make a choice, he’d probably grab whatever is closest, biggest, smelliest, or whatever he was last shown. That’s because dogs are unable to show any self-control when food is involved. That’s goes for all animal and small children as well. They grab something without taking time to contemplate the choice.

There was one nifty study in which a chimp was shown two different amounts of food, say 5 peanuts and 3 peanuts, and was asked which should be given to another chimp in the room. The other would be left for her. She always, always chose the larger amount first, even though she had to then watch the other chimp eat it. There was no doubt that she was able to understand the task, though, because this chimp had been trained to understood numbers. If the experimenter put numbers down instead of the actual food (the symbols 3 and 5), the chimp always, always chose the smaller number first. She understood the rules of the game but was unable to control herself to the point of pointing at the smaller amount of food.

Contrast Effects, Jackpots, and Reinforcer Sampling

If an animal expects a particular reward and gets something else, you ought to see his behavior change. For instance, a good reward is considered really excellent if the animal has just experienced a mediocre one, and likewise, a mediocre reward is considered pretty dismal if the animal has just had a good one. So if you normally use wieners for treats and suddenly one day, you present a piece of chicken liver, your dog ought to be thrilled and try even harder to get the goodies (this is called positive behavioral contrast). If you normally use liver but one day you discover you’re all out and have to use your dog’s regular kibble, don’t expect a resounding performance from your dog (this is called negative behavioral contrast)!.

We know this from research with mice running mazes. The indication of learning for maze-running is usually the speed with which the mouse reaches the goal box from the start box. If the hungry mouse always finds a piece of mouse kibble in the goal box (a decent reward for a hungry mouse but not terribly thrilling) and then one day discovers a piece of Stilton cheese in the goal box, the next time he’s given an opportunity to run the maze, he does so much more quickly than before. On the other hand, if the mouse is normally accustomed to finding Stilton cheese and one day there is only a piece of mouse kibble, the next time he’ll run the maze much more slowly than before.

However, it’s not just as simple as running hard for Stilton and not running so hard for kibble. The mouse that is accustomed to finding kibble but then finds Stilton actually runs faster than the mouse that always runs for Stilton. And the mouse that is accustomed to Stilton but then finds kibble actually runs slower than the mouse that always runs for kibble. In other words, to the Stilton-eating mouse, kibble is even more dreary than it is for the mouse that gets it all the time and to the kibble-eating mouse, Stilton is totally amazing, much more so than it is for the mouse that gets it all the time."

Reid, Pamela J. (2011-07-25). Excel-Erated Learning: Explaining in Plain English How Dogs Learn and How Best to Teach Them (Kindle Locations 702-714). James & Kenneth Publishers. Kindle Edition.
 

·
Registered
Joined
·
68 Posts
Key is to identify low and high value treats for your dog. Then start with the low value ones and save the high value ones for later when you encounter problems. Another way to look at it is, if you start with high value treats, then when you encounter problems down the road you have no other tools left (other than pressure).
 

·
Registered
Joined
·
3,722 Posts
Usually, learning is quicker the higher the value of the reward is. But, for complex tricks and chaining behaviour, you'd normally use medium value treats, and you'd use high value rewards for simple tricks.
 

·
Registered
Joined
·
348 Posts
Key is to identify low and high value treats for your dog. Then start with the low value ones and save the high value ones for later when you encounter problems. Another way to look at it is, if you start with high value treats, then when you encounter problems down the road you have no other tools left (other than pressure).
I sort of agree with this, although I'd elaborate that a truly effective trainer has everything the dog gets any sort of enjoyment out of (finds reinforcing) ranked somewhere, either on paper or in their heads.

So it's not just "high and low value" but like, scale of 1-10. Like, for some dogs, bananas are a 2. They're not very exciting and the dog needs to be pretty hungry to go after them, but they're still reinforcing. Kibble is a 4, store-bought treats are 5, access to water after a long, hot walk is an 8, hot-dogs with garlic powder are a 9, chasing the cat is an 11... etc. Then, as trainers, we use everything we can think of that a dog enjoys to influence behavior.

Sometimes we want a thoughtful dog that can work through problems. Sometimes we want a dog that is super excited and leaking energy. Sometimes we see a decrease in enthusiasm in a behavior and we want to "jumpstart" it. Sometimes the dog is SO excited about a behavior that there's no precision any more, and we need to make it less awesome.

I usually start with rewards that I would classify as a 6 or 7 to my dog and go up or down to get the interest level I need. Often, I'll use 3 or 4 "levels" of rewards in a single training session.
 
1 - 12 of 12 Posts
Top