New Idea: Continuous Memory Palaces

It’s been a few months since I created this thread, but there is something interesting to add.

It seems @Zoomy (Ben Pridmore) kinda uses this method already: he has commented on the forum that his loci work as a background, and that the images form a continuous story. This may imply that a continuous memory palace is efficient for memory competitions.

Since I have 3-digit PO lists now, it would be better to build a continuous chain of POP-O-POP-O-POP-O-… In other words, there is a person in the 1st locus using an object to interact with the 2nd person of the 1st locus, who uses an object to interact with the 1st person of the 2nd locus, who uses an object to interact with the 2nd person of the 2nd locus…

This POP-P strategy may be even more memorable than a POP because it’s often difficult to remember who the last person of the locus is. This is due to the fact that, in POP, the 1st person has two triggers: the action and its interaction with the object, while the 2nd person only has the latter.

For instance, if we need to build a scene with Deadpool (119) as the 1st person, a banana (922) as the object, and Robocop (497) as the 2nd person, we’ll probably imagine something that combines Deadpoool’s and the fruit’s essenses, but we won’t take Robocop into consideration to choose the action in order to be fast. I, for example, would imagine Deadpool shooting Robocop with a banana. What does this action tell me about Robocop? Nothing. I chose it exclusively because of the anti-hero’s humor and the shape of the fruit. The police officer’s trigger will be just my visualization of him getting shot, while the Deadpool and the banana will have both the sight and the action as clues. During recall, I may remember “There’s Deadpool shooting someone with a banana… who is that person?”

In a continuous memory palace, everyone, except for the 1st and last ones, has an action, so they actually have three triggers: interaction with the previous object, next action, and interaction with the next object.

In conclusion, this method probably improves accuracy and encodes 12 digits per locus instead of 9.

What do you guys think? Do you find it competitive?

1 Like

I’ve run into this exact issue recently. Been thinking about solutions for it as well. More from a cards perspective, but also for numbers. Prior to expanding to a 3-digit PO, my list was a mix of P’s and O’s so for numbers there was no way to guarantee the makeup of each loci or ability to make a consistent pattern. Now those doors are open!

An additional thing you could consider is the “impersonation technique.” Using your Deadpool - Banana - Robocop example, Deadpool uses the banana to shoot robocop, but the banana is dressed or decorated like robocop. Maybe it has a tiny little robocop helmet on it or instead of bullets it shoots little robocop helmets. I suppose this would only work well if the person that is being impersonated has a distinct outfit, but it could provide some additional clues to who that second person is.

I like this solution for numbers. Will give it a try. Although, I think I like to idea of just stretching the palace to have more loci and simplifying the scenes to just PO at each, then there’s no worry about a hazy second person in each scene, it only the primary subject. After all, a good familiar palace journey should be very easy to recall and is unchanging om each attempt, so making the journey longer shouldn’t be a big deal. With cards, you can’t guarantee a specific structure to your scenes. You may have several consecutive P’s or O’s. So, in that case, it may be better to have 26 loci with just a single item at each than having 13 loco but still dealing with PP scenes. Hmmm, I may be having a realtime lightbulb moment here… Gonna mess around with it today and will post more over on my thread so as not to hijack this one with it!

1 Like

Screw it, I’ll do a mini thread-hijack haha.

What I’m realizing may be that if your palace is very strong, the focus on “digits per loci” may be misplaced or irrelevant, especially when looking at it in terms of easy of accuracy during recall. If you have 4 extremely strong loci, you can fly though them very quickly and build a very simple scene at each with just a single element, either person or object, your choice, and encode your 12 digits across those loci. Having a single element at each forces interaction with the loci. Since you already will know the loci very well, when you recall you’ll likely be able to picture whatever the interaction was with the loci which should trigger the recall of the element that was stored there.

Now granted if you’re memorizing a very long sequence like 1000 numbers, then compression per loci would be very helpful because it could be difficult to prep and manage 333 loci for that, but lets say its for a memory league numbers competition where you have max 80 digits. This is 26 or 27 loci if you store a single 3-digit number at each one, depending on how you deal with the last 2 digits. This really isn’t that many loci. If you drill a few palaces so that they are really vivid and you can navigate them quickly, then it may be worth trying for an unambiguous single element at each!

I think the focus maybe should be less about trying to reduce the number of loci by putting more elements on each and more about using more and stronger loci so the scenes can be very quick to construct and simple to recall. At least, when it comes to competetive speed events. For a project like thousands of pi digits where there is no real speed factor, taking your time and creating vivid compressed scenes for many digits per loci is great.

Its kind of the old project management triangle of quality / speed / cost. If you want something done quickly thats high quality then it will be expensive. If want something cheap and high quality it will take a long time. Cheap and fast will be low quality. I think there a similar analogy here. You can have high data compression but the trade is higher scene complexity. You can have simple scenes, but the trade is needing more loci. You can have moderate levels of both. Its probably up to personal preference as to which end you prefer to increase, but its probably worth experimenting with all three approaches.

1 Like

Yes, I remember reading your post. The problem you faced together with Ben Pridmore’s comments motivated me to write on this strategy again.

It seems cool and creative, but that would kinda defeat my whole purpose of having extremely distinguished characters in my 3-digit list. I only took fictional characters because they have very unique combinations of traits, but if I mix them, then I may confuse one with another. Making the object impersonate the second person sounds more distinguishable, though.

hahaha, feel free to add ideas to it!

Yes, having a single image per locus is awesome for recall! I noticed that when applying VSE. Most of the memory palaces I memorized were very, very crowded, and when I placed just one element in each position, it felt so clean and sharp.

Exactly. My ultimate goal is the grandmaster title, but I’d need to memorize at least 1,400 digits in an hour just for the International master of memory title. Besides, I intend to break other memorization records of mathematical constants, which would require quite many loci. Therefore, having continuous memory palaces would help me enormously.

That’s a great idea. Now I’m wondering if I should apply this for short sequences and continuous memory palaces for longer disciplines or just focus on getting good at putting 12 digits per locus every time.

At 3 digits per loci, you’d need 467 loci and you’d have 7.7 seconds to work at each loci.

Compressing 12 digits at each, you’d need 117 loci, with just over 30 seconds to work on each loci.

Again, it comes down to which side of the equation you want to increase the workload of. Its tempting to have less loci and more time to work at each, but the simplicity of recall is also very appealing.

1 Like

I was thinking about @TheHumanTim 's comments, and I decided to test them by memorizing the first 648 digits of pi in six groups of 108. I tested three strategies:

  1. Continuous memory palace (12 digits per locus)
  2. Person interacting with an object (6 digits per locus)
  3. Person interacting with the locus using an object (6 digits per locus)

I didn’t plan beforehand, so these methods weren’t applied the same amount of times. I came up with the ideas during the process.

First, I tried to apply the continuous memory palace. I noticed a few things:

  1. If I forget an image, I may use the loci around to help me. This is really good for accuracy.
  2. This method doesn’t work well in convex corners because you can’t picture both loci at the same time.
  3. Because of the last point, you may want to cut the story at the end of each room; however, this makes the first image of the next room especially susceptible to forgetting.
  4. The images end up getting complex and generate a good deal of cognitive load to visualize and recall. This makes reviewing harder and slower, besides making you more tired.
  5. This method has a lot of story-making, so it feels fun.
  6. The story was fundamental to recall some of the loci fast (or at all). I have a long love/hate relationship with the story method associated with the memory palace: it feels fun and memorable at first, but all the systems I built that NEEDED the story became fiascos. Stories are the weakest links and can’t be reviewed, so they are forgotten after a while. If you can’t recall the loci smoothly without the story, this means you’re gonna have a hard time when they leave you. The clarity I felt was sometimes caused by the narrative behind the scenes. Without a plot, there are just a bunch of tortuous images.

Secondly, I tried to imagine just a person acting with an object (but not with the background). I noticed the following:

  1. I thought accuracy would increase because there were fewer images per locus, but it actually decreased. Having fewer images only made them more fragile because of the fewer triggers. (That is a big lesson for me since I thought fewer elements per locus always meant better recall)
  2. It’s difficult to make up a story for just two images, so it was just “convert digits into images and place at the locus”
  3. The cognitive load was much smaller, so I believe this method is likely to be less tiresome. In fact, reviewing loci in this case is quite simpler than in the previous one.

Thirdly, I tried to imagine a person interacting with the locus using an object. I noticed these things:

  1. This strategy had the best accuracy. That is probably because the trigger locus-image was reinforced. This seems to be the reason why the Variable Spatial Encoding of my 3-card system feels quite memorable: the image is ingrained in the locus. The point isn’t the number of images (the previous method was equally small but terrible).
  2. This strategy doesn’t allow much story-making either, so it feels less fun than the first one.
  3. Feels quite fast: convert and place.

I discovered a lot of cool stuff testing these strategies. One that I’d like to emphasize is that most of us have been comparing systems without taking cognitive load into consideration. It’s usually positively correlated with the complexity of the images, but pointing it out sheds light on why some strategies may be more tiresome than others and how long it will take for the accuracy to decrease in the long disciplines.

In conclusion, the only serious drawback of the third method is needing more loci; however, that is a one-time cost. The disadvantages of the first method, though, are unacceptable. Therefore, from now on, I’m gonna place 2 pegs per locus by imagining the person interacting with the locus using an object.

What do you guys think?

1 Like

Really interesting. The conclusions make a ton of sense.

This is something that I think is a big key. Involving the locus creates a strong link with something you KNOW is at the scene and can really help recall tremendously.

I wonder if there is an even better way to form these interactions. What if you tried to make the loci use the object on the person? Since you know what each loci will be in advance, you possibly could make the loci into living scenery and picture it starting the action of the scene. I wonder how this would compare to the Person uses object on loci approach.

1 Like

Yes, exactly! I had thought that just having fewer images would make the loci more memorable, but it seems that the number of elements isn’t that important, but rather the number of pegs associated with the locus. Alternatively, you may associate them with each other through a story, but these narrative chains are quite fragile.

Could you please give me an example? How would you imagine the locus acting on the person without making it become another element to be memorized?

I guess this would have the same effect since the number of locus-image triggers would remain the same. However, you gave me an idea: what if we imagined the locus interacting with the person, who uses an object to interact with the locus? This would connect both pegs to the palace!

1 Like

Hmm… Locus is a chair. Person is Lionel Messi. Object is a huge wooden Fork.

The chair’s arm animates and grabs the fork, hurling it at messi, who bicycle kicks it away.

Or, he kicks it back and the fork embeds itself into the chair, killing it. (I suppose this would kind of be that last idea you mentioned of loci affects person who also affects loci. I think that this is indeed the stronger scene.)

If your loci are objects themselves it could work well, as you could just anthropomorphize them. If the loci is less specific and more of just an area like “the living room” you could still do a similar thing, just make the whole room come to life and every object in the room grabs a fork and messi somehow kicks all of the forks back at once, killing everything.

This would be extremely easy to both encode and recall. Just knowing that the chair or the living room is the next loci, you can already see it coming to life and starting that scene going. And then the person is still strongly encoded because it reacts and re-interacts back with the locus. I don’t think it really turns the loci into an “element” in terms of an extra thing to add to the scene and thus adding workload to the scene creation, I think it takes advantage of whats already there by default and brings it to life in a simple way that kind of just depends on what object falls into its scene.

1 Like

I don’t want all of the scenes to turn into loci attacks person who counterattacks though… There would need to be other ways of dual action/interaction with the same object…

Chair skewers messi with fork, messi takes out fork and starts eating chair with it? I guess it would have to be very object dependent.

Say the object is a beach towel. The chair could be setting up a picnic by laying the towel out. Messi lays down on it then takes a towel of his own and playfully flick whips the chair with it…

The other consideration is I dunno if having to generate two interactions like that would cost too much time vs. just the singular “person uses object on loci.” I mean, it would have to, since there are more cause/effects happening. But it likely wouldnt be all that much, and could be worth it for the added strength of recall.

1 Like

This seems something a little bit different to me. It sounds like Locus-Object-Person-Object-Locus instead of Locus-Object-Person-Locus or Locus-Person-Object-Locus. What I mean is that you may end up forgetting Messi just like if we made a simple “person acting on the locus using an object.” You may remember that the chair kicked the fork to someone, who kicked it back, but Messi may be blurred in the middle of the scene.

An alternative to solve this may be to imagine the locus interacting with the person directly:

The chair holds Messi against his will, who stabs it with a fork to free himself.

However, this is starting to create a story… One of the big advantages of a PO approach is that it has absolutely no narrative. Although, it’s less fun, it’s likely faster. An alternative would be to, instead of imagining an interaction coming from the locus, visualize the contact between the person and the palace creatively:

Messi’s football shoe is stuck on the chair, so he’s trying to break it with a fork.

If his shoe is one of the things you pay attention the most when recalling him, then this would work well. In fact, I think that just imagining the contact between the person and the locus may aid during recall.

That would be interesting indeed.

Ok, fair enough.

I’m also thinking about that…

Now I gotta sleep. I may test these ideas tomorrow.

1 Like

As promised, today I memorized other 432 (4x108) digits of pi to test four systems:

  1. Locus-Object-Person-Object-Locus: imagine a personified locus interacting with the person using an object, who acts back using the same thing. (@TheHumanTim 's idea)
  2. Locus-Object-Person-Locus: imagine a personified locus interacting with the person using an object, who acts back without using the same thing.
  3. Person-Locus&Object: try to connect the person to the locus and to the object, independently. For instance, imagine the person rubbing against the wall at the same time they hit the floor with a hammer.
  4. Me-Object-Person: visualize yourself using the object to interact with the person.

I assessed each method using 4 criteria:

  1. Speed
  2. Cognitive load
  3. Accuracy
  4. Story (whether or not it naturally formed a story)

Here are the results:

1.Locus-object-person-object-Locus

  • Speed
    • It works, but it’s a little bit slow if you don’t want to make all images like “locus attacks image who attacks back.” In general, it isn’t as slow as I expected, so I believe that with practice it could become quite fast.
  • Cognitive Load
    • It was simple. The images aren’t complex at all. It’s ok.
  • Accuracy
    • The accuracy for people didn’t increase at all compared with a normal PO associated with the locus, so I don’t see any advantage in using this method. The person can easily get lost in the middle of the imagery. I believe this is because, in this technique, the first action is chosen according to the nature of the locus and that of the object, and the second one is imagined according to the nature of the first action. If you tried to choose an action that matched the person better, then I believe it would become a stronger trigger. However, this is quite difficult when you have 1,000 people who may share many features. In other words, any action would have really many possible agents.
  • Story
    • The images are complex enough to form stories naturally, but not to the point of needing the story. That’s good.
  • Additional Observations
    • I honestly didn’t like personifying the loci, but this is just a personal preference. Also, when I recalled the images a second time, I noticed that the people were indeed the weaker pegs.

2.Locus-object-person-Locus

  • Speed
    • Same as the previous one.
  • Cognitive Load
    • Same.
  • Accuracy
    • It was actually worse than the previous one, but I’m not sure why.
  • Story
    • Interestingly, I didn’t form stories naturally. This may explain the worse accuracy.
  • Additional Observations
    • I needed to stop at the end. I only managed to check the digits a few hours later. Yet, I noticed the loci were less memorable because I forgot more of them during the memorization itself.

3.Person-Locus&Object

  • Speed
    • Not very good. It takes time to think of a connection between the person and the locus, and they mostly end up being bad anyway.
  • Cognitive Load
    • It was okay.
  • Accuracy
    • This method managed to be even worse than the previous one. I guess that is because most connections between person and locus were just the person touching the palace. I needed to review the previous loci throughout the memorization.
  • Story
    • There was no story.
  • Additional Observations
    • I’m kinda tired now. In fact, I’m memorizing digits one hour before going to sleep. However, it’s much more silent at the moment. In the morning, I was memorizing while people around me were talking.

4.Me-Object-Person

  • Speed
    • It’s fast, but not as fast as a simple person interacting with the locus using an object.
  • Cognitive Load
    • Not much.
  • Accuracy
    • It was not as good as I expected, especially when the action didn’t match the person or the object. The visualization of the locus itself is not memorable.
  • Story
    • Sometimes there was a small justification behind the action, but most of the time it was just a generically weird action.
  • Additional Observations
    • Again, I’m tired now. However, I don’t think the bad result of this last system is due to that. I expected it to work better because the sense of touch would be involved and because of the diversity of angles. However, I concluded that visualization is of little importance. It’s just the means to create a logical connection between the locus and the peg.

In conclusion, I’m probably going to memorize numbers by imagining a person acting on the locus using an object. This seems to be the simplest, fastest, and most accurate method. It will require many loci, but that is a one-time cost. The only exception would be when I want groups of over 6 digits per locus. For instance, when memorizing phone numbers, I’d prefer to use a POPO method (the last person uses the object on the palace).

2 Likes

I was wondering if this would happen. Good to know. I think this method also just makes things too busy.

Sounds good!

1 Like

I believe my tests reinforce this conclusion. The best method was the one that associated most of the elements with the locus, instead of those that added more new information. For example, we may add stories to the scene, but they don’t last for very long and can be time-consuming; we may add more actions, but they only blur the elements and take time; we may imagine ourselves interacting with the pegs, which makes everything more vivid, but this takes longer and doesn’t improve retention.

In conclusion, encoding strategies that have fewer details between the locus and the pegs are faster, simpler, and more memorable. I believe this is due to the fact that these additional features add more content to be devised, visualized, and recalled.

1 Like

Tim, i only get more that 99% of objects correctly with repetition… without i don’t get this score… buy i’m slowly… 48 numbers in 5 minutes… many times triyng get 100% correctly… in 2021 finally get thus… @TheHumanTim

@AndersonAraujoBrasil please don’t take this the wrong way but thats quite a long time for that few digits considering how long you’ve been training. What encoding system are you using? If it’s 2-digits per image, that would only require 24 images to encode in 300 seconds. Taking more than 10 seconds per image is a lot of time. If you’re using a PAO, thats over 30 seconds to visualize each scene. This should be way more than enough to make things very memorable so that recall should be perfect pretty easily.

Have you been able to narrow down what your main challenge is? Are you not fast enough at translating the numbers to their images? Do you know your palaces well enough? Do your scenes not have a strong enough connection with your loci? Not to derail this thread, but we’d need some more specific info if you’d like us to try to help target some solutions for you.

Yes, @TheHumanTim… it’s very slow… but I’ve already tried PAO… I can’t memorize 100%… I’ve tried everything you can imagine, but to memorize everything, just with repetition… I’m trying to reduce the time, but I can’t. . I tried last year for a month without repeating… but I got 90%, 80% right… and only 20% faster than my method… I see memory athletes in names and faces getting 60 names in 5 minutes and 60 words in 5 minutes… my record was 24… so I think, I think I’ll continue like this and try to improve… when I started in names and faces it was 14 on average, today it’s 20… so if elite memory athletes get 60 names and faces, 80 words… both of us are good and I’m slow at both… so I’ll continue training like this, who knows, maybe I’ll increase it to 40 names and faces and 35 words… because these athletes are good in all disciplines. . I’ve tried everything, but only in my own way, I can get 100% in 15 minutes, 10 cards… I think, I try to find a faster way, but I can’t… I’m training and I’m looking to find out… like Alex Muellen, I saw in an interview that he has athlete secrets… I’m trying to improve… @TheHumanTim

So, since I started training in the memory championship disciplines, I wanted to get 100% of the objects right… I memorized and I always couldn’t get 100%… or 60%… so I tried… until I saw on an app that many athletes from memory they repeat… including Ron White in his course (very good course) he says there are 5 locations in a department… memorizes three, revises, then memorizes the other two and revises… for me, it only worked… (more 99% revising, and trying so many techniques, with repetition I get a good score… it took months of trying, then (like a game of 7 errors) I questioned myself why I memorized one object and the other didn’t… but like that, until In the 5-minute names and faces, I memorize in 3:40 minutes and the other 1:20 I review… for now without repeating… I wasn’t successful… I don’t know if the top memory athletes don’t tell their secret. What I have improved in the last few months is faces and names… my average was 14, but with repetition and feeling my body telling me that it is memorized… today I can get 21, 22 faces… I only got it like that… I think I’m going to continue and try to be faster… @TheHumanTim… I’m betting that I can improve the brand… since I started at IAM software in a year, I’ve improved about 20%… I use 000-999 for numbers… after April I’m going to learn two cards in 1 image… that’s it, sometimes I think of a faster method, but I can’t get it to 99% anymore… for example, in my training this week, I managed to add up about 700 objects and only 1 error that When I tried to find out why I made a mistake, it’s because I didn’t do the review properly… but with the review it’s practically certain that you’ll get it right and memorize everything… @TheHumanTim

Ok, this is interesting. With a 3-digit system, to memorize 48 digits you only need to encode 16 images. If you can practice your recognition speed and get it to two seconds or under from the time you see the number to the time you picture your image, that should only eat up about 30 seconds of your time. If you take around one second to decide how to place each of these images at their loci, you’ll be at under 50 seconds. You’d still have over 4 minutes to review if you think you need it. Thats enough time to review your 16 images many times and should be plenty to lock it in so that you can recall them quickly and accurately. If you can’t associate the numbers to their images that quickly, I’d suggest working at random number recognition so that it becomes automatic. This will let you totally focus on making your scenes vivid and memorable. Good luck!!

@TheHumanTim Yes, until last year I used 2 digits per image… end of last year and January, I already have the exact 999 number… when I memorized 2 digits it was really fast… 95 (ball)… today 997 (small baby) … as I’ve only been using the 999 images for a few months, it’s a bit slow, but it’s progressing little by little… like, in each department there are 4 places to put the objects… I memorize all 4… then without looking I review them without looking. . in terms of correct answers, everything is ok… I’m betting that I can become faster over time (like a computer keyboard) the first time we saw someone typing quickly, I never imagined that I could also do it with practice… let’s see how my performance goes… I memorize focused on 100% of the correct objects… thank God I already managed to do it…

1 Like