N-back factors and your opinion on them

[Fredo Corleone]

Hi guys,

I would like to know your opinion on the different N-back factors and what you think they stress and how you tweak them.

Following I've listed the unique factors I've discovered while playing the game:
- Number of dimension (single, dual, quad, ...);
- Stimulus duration (how long the stimulus stays on);
- Delay between stimuli;
- N level;
- Matching stimuli density (matching stimuli / total stimuli);
- Tricky patterns (ex. for N = 5: A B C D E, B C D E A or A A B C D, A A A B C);

- Repeated symbols vs row of different symbols (ex. for N = 5:  A A A A A vs A B C D E);
- Stimuli set size (ex. 4x4, 9x9, 16x16 grid so on);
- Difference between individual stimuli (ex. sounding of C and T is further than T and K).

I feel everything plays a role on adding or subtracting cognitive load to the exercise.

For instance I feel immense relief when I play higher N levels and I hear/see the same symbol for 2-3 stimuli in a row, there's surely some compression my brain is doing and it feels easier.

Tricky patterns add cognitive load for instance when I hear letters that are similar in sounding and they are inverted between a sequence of N and the next one, or when the next sequence is simply shifted by one stimuli and it feels like a déjà vu.

Another thing that adds massive load to cognition for me is the set size, infact with smaller grid or fewer letters there is more interference and I'm constantly trying to filter out misleading patterns.

Let me know what you think.

Fredo

[Fredo Corleone]

Ah, I forgot to mention style of play, whether you chunk or use some kind of rehearsal technique (I don't).

[Leonardo]

The best n-backs I have ever played are your v0 and v2. v0 for how it stresses your spatial vision, I don't know of any software that do this better, I would only hope for more shapes, like a sphere, a pyramid, etc. v2 for the sheer amount of stimuli flying in front of you. I have never been able to go beyond 9-2-back. 

[Leonardo]

Something that I really like, btw, are modes that force you to manipulate what you have memorized. For example, arithmetic like in PASAT. You already did a PASAT app. v0 with different shapes and number manipulation would be a really interesting way of potentiating spatial vision. 

Spatial ability and mathematics are closely related, so it makes sense to train them together:

  Previous research has established a link between spatial ability and mathematics —children and adults who perform better on spatial tasks also perform better on tests of mathematical ability (Burnett, Lane, & Dratt, 1979; Casey, Nuttall, & Pezaris, 2001; Delgado & Prieto, 2004; Geary, Hoard, Byrd-Craven, Nugent, & Numtee, 2007; Holmes, Adams, & Hamilton, 2008; Kyttälä, Aunio, Lehto, Van Luit, & Hautamaki, 2003; Lubinski & Benbow, 1992; McKenzie, Bull, & Gray, 2003; Mclean & Hitch, 1999; Rasmussen & Bisanz, 2005). This link may be based on shared underlying processes. Brain imaging studies confirm that similar areas are activated when people process both spatial and number tasks (See Hubbard et al., 2005 and Umiltà, Priftis, & Zorzi, 2009 for reviews). There also is behavioral evidence that the two are connected. For example, studies indicate that number is mentally represented in several spatial formats (e.g., the SNARC effect, object files, etc.) (See Mix & Cheng, 2012, for a review). The connection between space and math is so compelling that many now believe spatial training could be an important resource for improving performance in STEM disciplines (Lubinksi, 2010; Newcombe, 2010; Uttal, Meadow, Tipton, Hand, Alden, Warren, & Newcombe, under review). In fact, the National Council of Teachers of Mathematics now recommends integrating spatial reasoning into the elementary mathematics curriculum (NCTM, 2010).

https://files.eric.ed.gov/fulltext/ED580905.pdf  

[Leonardo]

You can also make a 3D n-back with number manipulation and analogy. Does this two stimuli share the same relation than this two stimuli? We are on the absolute infancy of mental training. In the future, and with the help of VR, one will be able to increase his IQ at will.  

[Fredo Corleone]

The next thing I would like to make is a Syllogimous v4 with procedural generation of Advanved Progressive Matrices.

The composition approach you suggested is very similar to a formula that involves APM.

I would like to talk with you about this in a Discord call, let me know when you are free!

[Bo T]

If i recall correctly, the latest meta-analysis on N-back training has shown that dnb had an effect size of 0.24 on generalized working memory (working memory tasks different than the trained task). Do you guys think that different modes might yield different results?

--
You received this message because you are subscribed to the Google Groups "Dual N-Back, Brain Training & Intelligence" group.
To unsubscribe from this group and stop receiving emails from it, send an email to brain-trainin...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/brain-training/0e98eb37-f4c3-4db7-bc06-23ef7ad64810n%40googlegroups.com.

[Fredo Corleone]

I believe N-back has transfer under the following conditions:
1. The player doesn't strategize, no chunking and no rehearsal, and obviously no mnemonics;
2. The player train long enough, often enough. As with anything improvements follow a curve of diminishing return, the more you play the less you gain, that said on average you are expected to gain N=0.25 every hour of play with a level up threshold of 70% of correct answers. The stricter the level up threshold is the less your expected N/hour gets;

3. When you test for transfer you keep in mind there are different neural circuits involved in different tasks. If you only train your phonological loop (audio N-back) that won't transfer to visuospatial tasks (matching positions on a grid). I believe that if your audio N-back gets higher than your digit span, then your digit span will increase too.

[Leonardo]

I know that you have been thinking about procedurally generated matrices for a while. That's a nice concept, but I will believe in it when I see it working properly. 

My phone is literally this: 

That is why it is not easy for me to have conversations. 

[Fredo Corleone]

Well, I meant having a call on Discord. No need to bring your Alcatel to the table xD

[Leonardo]

I mean. I can't install Discord on my phone, because it doesn't run apps, and I don't have a microphone for my PC. That's the reason why I brought up what kind of phone I have. I cannot use Discord with it. 

[Grey Bach]

I do quad n-back with 12.5 interference currently, because this mode stresses all the stimulus types in the game. I do the default 3 secs, but think I could benefit more from doing 3.5-4.5 seconds given that I have relatively slower processing speed. I suspect the optimal for this will be variable for everyone, but in my mind there is a processing stage where you are still trying to adequately encode the sequence and then there is the holding phase where you simply have to hold in your mind what you have encoded for a certain amount of time. I think you want the intervals to be long enough that you are able to train both. If you go shorter than the processing phase, I think you begin training processing speed at the expense of wm. For the stimulus duration, I'm not sure what would be optimal. I would assume that it follows phases similar to delay between stimuli though.

As for number of dimensions vs n level, I am not certain whether it would be better to practice simultaneous stimuli at lower N levels or just one of each at the higher N levels. I have tried some simultaneous and noticed that it tended to tax my processing speed and attention to detail far more than my wm. Perhaps this would be different with longer time intervals or longer stimulus duration though. At that point, it is just a matter of unevenly timed stimuli that give you more flexibility in the order you process them (multi stim) or evenly timed stimuli where the order you choose to process is more predetermined (single stim).

I've been thinking a lot lately about the difference between what we might call wm volume and wm manipulation ability. Wm volume denoting the theoretical number of storage slots you have in your head when presented with 0 interference/similar stimuli and each stimulus is a simple as possible (a fundamental unit). Manipulation ability would be ones capacity to use ones wm volume toward some end in the face of distractions/interference. Digit span forward would be more a test of wm volume whereas digit span back and seq would test more wm manipulation ability. I think it would be cool to explore how training varying degrees of each transfers across the spectrum. How much better would one get at digit span back if they only trained digit span forward for example. Similarly, I plan on importing more colors images and sounds into my n-back game and perhaps changing the interference level to 0 to try and train my volume as much as possible for a while to see how much it transfers to my manipulation. I also think playing on a larger grid could be interesting.

[Fredo Corleone]

I personally train audio N-back (so a single dimension) with 1s between stimuli, I think that playing that fast prevents the use of techniques. I would like to train at 750ms but I have a problem with bluetooth headset delay as it leaves too lil time to match stimuli.

Between single and dual N-back there isn't much difference for me, my N level is limited by the auditory part.

[Leonardo]

I play at dual-5-back on your v0. There, the audio is really easy, the limiting factor is to keep track of all the relative positions when the cube spins quickly. Don't you train with that one? Is hands down the best software for this kind of training by a huge margin in my experience. 

[Fredo Corleone]

From time to time just to break monotony, don't really train with that because it's mind bending.

[Leonardo]

Is not to bend our minds what we want? 

[Leonardo]

I think that your v0 is responsible for at least part of my ability to draw like Kim Jung Gi:

[Leonardo]

Imagine what I would be able to draw if you implement other shapes! 

[Fredo Corleone]

What else can be added? Maybe playing with perspective: showing a cube but from different PoV and perspective values

[Leonardo]

The issue I have with that is the one that I have with v2. That it is really hard to track the simultaneous movement of the cube and the camera, sometimes an optic illusion makes you believe that the cube hadn't move at all when it is not the case. 

I would be happy with just more shapes. I don't want to end up over-complicating and feature creeping this app. 

Of course, I would made the shapes transparent and made up by pieces, the same that the current cube is. Exactly like it is right now, but with more shapes. 

Optional arithmetic options, for the reasons stated above, and maybe color may also be interesting for space ability. 

[Leonardo]

A couple of self-portraits. We can track my ability to draw from imagination. Lol.

[Leonardo]

[Grey Bach]

Has audio only improved your performance more in dual than training dual directly, Fredo? I have considered training my weaker points separately in quad as well.

[Fredo Corleone]

@Leonardo nice drawings! :) About the shapes I have to think about...

@greyson I think it's improving my performance on dual, I will tell you when I'll be capable of consistently playing at N=7

[Leonardo]

I hope that you find it useful. More shapes would greatly improve visualization and motivation. Dealing always with the same shape is the opposite of cognitive flexibility, which is the main aspect we should be training. 

Intelligence IS Cognitive Flexibility

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9397005/

I think that you should give a chance to your own app, no matter how deeply your mind bends, it is really good. 

[Leonardo]

As showed before, training with a diverse array of 3D shapes would greatly help you mitigate this issue:

Recently I started playing Magic The Gathering, which is a collectible card game, and my arithmetic weakness is showing off pretty bad: I can't estimate points of damage quickly, and I have hard time figuring out how many cards are on the board,  even when there aren't that many. And I have to ask my opponent for out many cards they have in hand even if I'm able to see them perfectly.

Training spatial cognition enhances mathematical learning in a randomized study of 17,000 children

https://www.nature.com/articles/s41562-021-01118-4

[Leonardo]

Btw, the stimulus bar seems to not work at all: 

It would be nice to be able to make stimulus shorter and rotation faster. I sometimes get bored at the slow pace of the software and I start thinking about other things. 

[Leonardo]

I have never been able to beat N=5, although I have came close a couple of times. But I beat N=4 regularly. The leap is quite significant from 4 to 5. 4 offers a little challenge, but it is too slow. I fail more often by getting distracted while waiting for the next stimulus than by my memory. I find surprising how much negative stuff comes to my mind while training. Every single humiliation I have ever suffered comes rushing to my memory when I'm trying to concentrate. 

[Fredo Corleone]

Pushed the fix ;)

[Fredo Corleone]

Back in the days N = 4 was not hard at all for me, but I didn't play with rotation on.

[Leonardo]

Thank you, Fredo!

Without rotation I can do double easily. The rotation is the meat of the training, in my opinion. 

[Fredo Corleone]

Adding different shapes in 3D Dual N-back would make it Triple, do you want the sound to be removed instead and the second stimulus to be the shape?

[Leonardo]

I wasn't referring to that. In V0 you have a cube wireframe, but the cube wireframe is not part of what you are asked about, you are asked about the lighten position inside the cube  wireframe, the cube wireframe is just the "territory". Every game you could have one different "territory" chosen at random: a pyramid wireframe, a toroid wireframe, a cylinder wireframe..., instead of the cube wireframe, but performing the same exact function as a "territory". Just so you adapt to different shapes, and really develop spatial vision, instead of strategizing your way up. 

Also, variable mode, as seen in Brain Workshop, where you are asked to recall any number of steps up to your N level. If you are at 5-back and are shown a 2, you must remember as in 2-back, etc., wouldn't be considered a third element either. 

Variable N-Back
For an extra challenge, try enabling Variable N-Back by pressing V in the Choose Game Mode screen. Each trial, a random n-back level will be displayed in the center.

https://brainworkshop.sourceforge.net/details.html

I think that different shapes is essential for developing spatial ability and variable N-Back is essential for developing working memory. I'm pretty convinced that the reason why we don't see transference after most training is because we just become better at a very narrow set of skills and strategies. The more diverse, the harder to strategize, and thus the better chance to train the underlying construct we are trying to potentiate.

[Leonardo]

 As expected, both training groups showed improvements in task specific working memory performance and reaction times. However, there were no transfer or novelty effects on fluid intelligence, verbal memory, digit-span, and executive functions. At the neural level, no significant micro- or macrostructural changes emerged in either group. Our findings suggest that working memory training in healthy older adults is associated with task-specific improvements, but these gains do not transfer to other cognitive domains, and it does not lead to structural brain changes.

https://onlinelibrary.wiley.com/doi/10.1002/hbm.24965

Regular working memory training has failed again and again. I think that it may be useful, but not without variability. 

[Leonardo]

If you imagine a human brain as a neural network, you can easily understand that, by extensibility training with a program that shows no variability, and therefore have a very limited set of elements, we are just overfitting. No real learning is done. That is why I'm obsessed with cognitive flexibility and constant changes of rules. This is the only way to avoid overfitting in our training. 

I think we have reach a point in which overfitting is our main issue and the reason why we don't become substantially smarter.