Nagime memory training begins

As people may already know me on the forum I have had relatively a lot of success with memory techniques being applied to what I do however in the past year or two I have spent most of my time with rote and really dragged the dust out of it. I have made my ability to rote(of course with spaced repetition) learn effective enough to be usable, the issue remains that it is slow to learn with regardless. I can reliably handle a working memory load a minute with rote which may cap around 2000-3000 words(associativity) in 10 hours if we are talking about learning words. My memory with rote has increased drastically in high periods so I can definitely see it improving further but the core of it is that it isn’t as fast to memorize with as memory techniques are. 150 minutes which is 2 hrs and 30 minutes can yield 3000 words with memory techniques that is at-least 4x faster, furthermore for foreign alphabets according to my own tests it is incomparable, you can just memorize an entire table with memory techniques in a go without having to apply much spacing at all and it sticks very well.

As such this is an official restart for me to dive directly into memory techniques while it won’t entirely eliminate my rote training as this has taught me a lot about the brain, I will focus far more on memory techniques once more.

While I am not planning to enter competitions I will aim for the highest output, but I will naturally be inclined on memorizing words as it is more likely to directly fall in line with my learning and also improve speed.

For restarting memory techniques I will set some simple initial goals and extend on them:

Make 1000 new loci
Make a full 2 digit major system
Make a PAO system perhaps for cards
Make a radical system (kanji)
Solve the issue with fast chaining
(this is where you slow down when you try using already learned images quickly and interchangeably) I hope to solve this without having to learn the chained pattern but that is always an already known solution.
Large experiment on properties to optimize memory techniques (fun)

I will try updating at-least once every week and keep a regular schedule to work on this.

5 Likes

I have been internally debating how to acquire the appropriate systems but this might as well be a chance to use memory palaces to learn the systems though it would be easy enough to use rote.

First task -
100 loci and encode 100 major system digits inside them.
check a list of numbers and add the sluggish digits to a more frequent review - test some review schedules on the list and the sluggish numbers.

Notes -

  • Avoid checking the list in order and check it by random unique numbers -> simply write a script to randomize some numbers from 0-100 and print them in a 5 second fashion, note the digits that are slow.

This only leaves whether to write down the locations or learn to keep them under review so that I won’t have to. For the sake of the initial start, writing the locations down will be done.

I will also test minimized versions of loci to see if I can squeeze in locations without decreasing potential use.

3 Likes

Things were a bit delayed unexpectedly , this provides evidence for me to rework my scheduling a bit.

The initial 100 loci were created on the first day in a surprisingly short amount of total time around (15 minutes). I have then maintained them over the week and attempted to place some images out of the major system but I was getting reluctant with the lettering.
10-19
| /
V
t/d[aiueoyh]{s/z,t/d,n,m,r,l,ch,k/g,ph/f/v,p/b]
I should have definitely used an existing major list or generated one rather than making one from scratch. It is much easier with larger numbers because of the variety and length of words, but when you start having to use words like teas,deed,tan. It honestly feels faster to just make a peg list system. It also seems more compatible with me because I can think of the images dynamically much faster without restraints.

Mid using the major system while it was more sluggish to begin with, the next day (even though I hadn’t generated a lot of images) I was at-least twice as fast. This was for committing the encoding and thinking of the similar words. For larger numbers the major system is definitely more effective than for 2 digit samples.

The obvious issue came to attention -> I may know my 17,18,19 in my journey but I do not have an idea of which is which unless I counted them or dissected the name. This is problematic if I need to store my images very quickly, as I would have to use the major system much more quickly and then wouldn’t really benefit from making images at all. Overall a slower process for now.

This led to 3 very obvious options:

  1. I encode the digit as I see it along with the image. Teas (pair of tea bags) can for example have a written 1 and a 0 to indicate 10. I know this works well from previous experiments since it lets you instantly think about the images as you see them
  2. I use a shaper system, which acts kind of like #1 but does so in the reverse order.
  3. I use a link methodology.

Crucial importance of this: In general when you are memorizing things it doesn’t matter how structured the memory is, sequential order is good or often essential for recall. For scientific/mathematical information, the links between things are so much more important that a sequential order doesn’t provide any benefit over a textbook. Hence these 3 are very good considerations. As this is ideally applied to information as well my bias to the methods will be #3. Further justification for #3 is due to theory being visually distributed as well, for example in mathematics you can generally recognize a visual representation for theory.

In order to gain the benefits of memory, the memory palace can still be effectively integrated ~ some experimentation on this is underway but the most obvious and likely least useful method would be to structure the memory palace itself. Definitely doable as I am making my own memory palaces but it would restrict the speed a little as it isn’t as free.

The current simplest way for further finishing my list would be to simply anki along the things I need to remember but for the sake of experimenting I will do it in a mnemonic compatible way.

Overall my first priority will be reworking on my scheduling a bit so that I can stay rather more consistent in the limited time I have. Following this I will revise a better link methodology after testing the common cases e.g new version of memory palace -> link palace, theoretic imagery to information, super peg chains (a way to link imagery and maintain abstract meaning).

The obvious consideration will be the stimuli, it is kind of easy to dissect how I have to make this work based on the task. In the case of memorizing numbers, quite simply the cue is the visual sight of the numbers. Whatever I do there must be a mnemonic way to perceive this cue to apply mnemonic techniques. Essentially though for the sake of more advanced techniques, I will attempt to ensure the processing of the images is more memorable than just calling the number 102 a boat on a rock. The fundamental tools for this are sound,sight,color and movement(be it size or magnitude and direction).

As for the minimized versions of the loci, this worked well when the space I was in was furnished as there was a lot of variations. A visualized beach is kind of too similar to serve as more than 3 loci if you include the split to the sea. On the other hand my 1 loci knight room instantly turned into a 20 loci room with some minimizing.

1 Like

I went a little crazy on the testing and my notes alone are over 10 pages but I will post a broad summary of some changes to keep the post clear.

After a lot of trailing and error I have come to the consideration of using my custom morph system. My morph system a little which is similar to the shaper system but instead you mentally convert the text/numbers you are seeing into images even if you need to distort them. The reason for this change is because there is no training time in using a system which takes the cue as the input data to produce an image. The shaper system is also equally compatible. The reason for me choosing the morph system over the shaper system in tests was as follows :

  • It takes me less time to make images as I don’t have to think of suiting images but instead morph the text to be suiting to any image.
  • My tests showed that if I encoded 4 digit numbers in real time with the morph system I didn’t lose track of numbers. If I do the same with the shaper system because the shape of the image in the moment isn’t precise enough (at-least for 4-5 digits) I lose digits in real time. It seems with the morph system the detail from the encoding of all digits as one is retained.
  • It retains the same benefit of requiring less training to be instantaneous as the shaper system.

Alongside this I also tested :

  1. Normal shaper

  2. Encoding the writing of digits on other images with memory palace
    (more memorable than expected but not significant).

  3. Placing morphed images in a memory palace
    (seemed less effective for speed but was good for general storage)

  4. Morphing shaper images rather than the digits.
    (Strangely there wasn’t any confusion of using the shaper morphed images first and it improved recall but not significantly)

  5. Normal Morph
    ( had higher recall than initial shaper but was 0.1 seconds slower per digit)

  6. Link method+sharper
    ( took 8-9x the time of normal shaper but had perfect recall besides 1 error)

  7. Link method+morph
    (took 8-9x the time of normal morph but had perfect recall)

  8. Normal sub-vocalization.
    (Didn’t do well up to 20 digits but was 2.5 times faster than shaper).

  9. Spatial system
    (It felt strange but was unique and okay effective took 2x the time of the shaper system)

  10. Spatial link system
    ( took 3x the time of the spatial system without it)

  11. Combined spatial and shaper system
    (while slower than the spatial system 1.5x, interestingly with the first time using mnemonics I was aware of the quantity of each digit in the sequence, even if I did not remember their order so well )

  12. sound system
    ( interestingly took 0.06 seconds more than the shaper system and was less flexible)

  13. Shaper in memory palace
    (took around 3x normal shaper, recall was perfect but notably speed and recall was faster if the digits were closer to assume a chunk than if spread out across the loci).

I applied all of these on 1 digit systems. My rules did not permit me to use chunks. So I was evoking the recall of 1 image and then moving to the next. I realized quite a lot of things in doing this and I have the exact data as well but it would make my post overly long.

Firstly a spatial system as I defined it is essentially similar to a PAO system but it only encodes the act of movement. I was kind of curious as I knew the brain region for this was heftily involved in memory.
When I did not combine it with the image I used a random image to produce the movements, in my case I simply visualized a blade and random persons swinging it in shape per digit. As it was overly unique of an experience ( it clearly felt different and had a hint of some more things to discover ), I decided to further test on spatial things in the future.

The sound system encoded a shorter sound per digit that I was not used to using. It was slower than the sharper system which is a good indication that the sharper system is fast enough out of the box. Regardless it was very obvious that where I was losing all the time was not in recalling the digit but moving onto the next digit after recalling the previous one. There was some kind of refractory period even if I clearly knew which image was next. This also applied to all systems but was pruned on the spatial systems.

I got to a number of conclusions :

  • simple solution 1: learn chunks

  • simple solution 2: parallel encode

Firstly these methods would obviously work. By learning chunks you would learn how to quickly place digits after the others. The downside to this is that it takes a long time to cover all possible chunks, far more than making another 5 digit system when you get to a 3 digit one.
Parallel encoding would naturally be encoding multiple digits at a time so it would have more chunks in a single go. The downside was that this itself required training and wasn’t exempt after hitting the adequate chunks, in the end it would also suffer from fighting peripheral vision. Also, most of all, memory champions who train a lot likely do not use these 2 principles but still get their time down significantly.

As a result of this I considered really what other possibilities there were. Theoretically ‘having a stronger connection’, ‘more efficient synapse’, while it makes sense is kind of on the wrong point. What I was facing was a refractory period which either of these isn’t expected to help much with, and for a synapse learning a chunk would naturally be effective. I thought there should be another thing that can be done besides having a upgraded synapse, cross activation’s and ‘stronger connection’. The shaper system essentially has a very strong connection to the digit because it is self contained, the connection should be efficient. When I considered this it really hit me though, what I was essentially doing was evoking 1 image, stopping it and then starting the next. It’s a very logical thing we don’t usually overlap images when evoking 1 at a time but I realized 2 things here:

  1. I was artificially waiting for the first image to fade before I started integrating the next image.
  2. When I tested the spatial system It felt very different because I was not really waiting for anything to fade. This point is really logical because we can draw on paper we don’t have to wait until a line is drawn and say okay before we draw the next one. More so we don’t have to do this for every mark the pen makes on the paper.

Considering those 2 things I realized some precise ways I can reduce the time otherwise.

  • Increase the speed at which I clear the image
  • Place or activate an image in a way that I could equally so and easily so activate a second or third so I don’t have to wait for it to end.

To my surprise or perhaps as expected, increasing the speed at which i clear the image led to increased encoding speed and reduced time per digit proportionately to how quickly I can clear the image. Even more surprisingly when I started the process of converting images to numbers while staring at the numbers in a spatial field like visualizing the images on the screen as I encode them with my eyes open in mini version. My encoding speed essentially skyrocketed to some competitive levels. Even using a memory palace in this way was much faster. This is also for 1 digit precisely to make testing more effective. The results would be at-least 3x faster with a 3 digit system, but I could definitely encode the digits doing this faster than I could read them.

I will plan to do some speed reality checks when I extend my system, for now the 101 image morph system is complete, for digits 0-100.

Essentially I have indirectly solved the issue of fast chaining, which I will verify when I fully complete few systems. I believe it is more effective to a separate post on updated goals so I will do so shortly.

Since I have validated my decision on the system I will use I will now also apply spaced repetition as appropriate (I reintroduce anki here). With the memory capabilities of shaper and morph systems, I will likely be able to work in batches of 100 digits comfortably.

Further tests will include a structured memory palace for information but this will be post the next goals (I hope).

2 Likes

Update on Goals:
Reintroducing Anki

Micro-schedule all tasks
Make 900 more loci
Make 100 structured loci
Extend the Morph system to 1000 digits
Make a PAO system perhaps for cards
Make a radical system (Kanji)
(Solved) solution to the issue with fast chaining
Large experiment on properties to optimize memory techniques (fun)

1 Like