Which is better, triple nback or dual nback?
Tim
at your limit, or if playing tripple nback might actually increase
your gains more rapidly.
Gwern Branwen
Hash: SHA512
Well, the usual idea is to play at your limit; if you are playing dual
5-back when you can handle triple 5-back, then obviously you should be
doing triple. But if the choice is between dual 7-back and triple
5-back (or something like that), then I don't know of any reason to
prefer one to the other.
There is the general observation that 'if it's good to go from single
n-back to dual n-back, then it'll also be good to go from dual n-back
to triple n-back', but who knows whether that's true? (I can't speak
from any personal experience.)
There is one known way to gain more rapidly besides playing a lot at
your limits: practice just before you go to bed. If you want to read
about that, I uploaded the paper on this to the group's files.
- --
gwern
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
iEYEAREKAAYFAkoIruYACgkQvpDo5Pfl1oKnFACfd5RH/VXfByopIFcPHSTwbriB
kz4AnjcMNB+jNWclHD8KNSB2WagFewQU
=BHY/
-----END PGP SIGNATURE-----
Pheonoxia
seen any of the files Gwern has uploaded, you'll know those can be
trained in accordance with working memory.
Triple allows you to simultaneously memorize 50% more stimuli. With
regards to a possible transfer effect, I believe mastering triple
would allow one to hear and account for things without actively
listening. For example, someone is talking to you, but you're paying
attention to two other things, say a moving object and/or something
shiny. Through training with triple, you'd be better able to engage in
what they're saying while still focusing elsewhere.
That being said, I'm horrible at triple, largely because I've mostly
trained in double. But now I think I'm going to train lots of triple.
Denis Gorodetskiy
> That being said, I'm horrible at triple, largely because I've mostly
> trained in double. But now I think I'm going to train lots of triple.
I'm getting about 70% at visual and audio and about 40% at color while
playing triple n-back...
BTW guys, do you think it would be cool to develop quad-n-back?
I suppose fourth axis could be using geometrical 2-d shapes or 3-d
models or something.
Gwern Branwen
Hash: SHA512
On Tue, May 12, 2009 at 7:54 PM, Denis Gorodetskiy wrote:
> BTW guys, do you think it would be cool to develop quad-n-back?
> I suppose fourth axis could be using geometrical 2-d shapes or 3-d
> models or something.
In a facetious vein, I suggest for quad n-back the following scheme:
One is presented a 3x3x3 cube (visualized in 3D of course!); one
modality is letter; one modality is color; one modality is x-y-z
position. For quad, they merely have to be on the same x; for
quintuple n-back, x or y-axis; for sextuple n-back: x, y, or z axis.
For those wishing octuple n-back, the cube obviously generalizes to hypercubes.
- --
gwern
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
iEYEAREKAAYFAkoKDeQACgkQvpDo5Pfl1oJk2ACfVOS/aYoLP+1aY3NGMyQ8/1i5
Z4sAnAwoUdN2Ygy2jLTGETnTV/6mHLcO
=ptFR
-----END PGP SIGNATURE-----
negatron
> That being said, I'm horrible at triple, largely because I've mostly
> trained in double. But now I think I'm going to train lots of triple.
color alone and you will see what I mean.
The brain apparently has distinct processing areas for color.
The problem with a high simple n-back level is that eventually
retention could be the limiting factor rather than capacity. Retention
is perhaps not as trainable if at all.
Moving to a more complex n-back allows to cram more items in a shorter
period of time, and if color has anything to do with it, seemingly in
a different region of the brain, which might be necessary for
advancement eventually.
I can get over 90% consistently on dual 7-back, however I've been
having difficulty moving from 8-back for a month. I now moved to
triple-back but it turns out that because of color I have to start
from the ground up. I have to focus on color fairly strongly with
mediocre results at 3-back at the moment. Position and sound come
through practically 100% without paying any explicit attention to them
at this point. I could exclusively draw my attention to color and not
miss position/sound, nevertheless color performance is embarrassingly
bad.
Pheonoxia
haven't trained much with it. I try to focus exclusively on color and
the other two come subconsciously. I'm just starting to get T3B down
well, so hopefully I'll be doing T4B by the end of this week.
I've noticed two apparent strategies for memorizing color. I'm opt for
the second so my verbal memory remains independent of the process.
1) Say the name of the color in my head
2) Visually memorize the color in my head based on its location, which
is harder (for now)
Do you have any evidence that the brain processes color separately
from visual locations? Or is it just your experience?
negatron
habit. I realized that doing so was training something else (not quite
sure what) rather than inherent ability to retain color. I wonder if
mentally verbalizing an item to remember is a close equivalent to
having the item spoken out, in which case you could be translating the
problem to auditory working memory, which is already being trained.
> Do you have any evidence that the brain processes color separately
> from visual locations? Or is it just your experience?
memory should attain equivalent n-back levels, but for some reason it
does not.
negatron
two.
Position n-back is 2-dimensional visual
Auditory n-back is 1-dimensional
Color n-back is 1-dimensional visual.
Perhaps it's not because it's color, but because there is no planar
separation of elements that is the problem. If the color was replaced
with a shape, it's possible the results would be just as poor.
It may be that we are tapping into another important working memory
process that dual n-back doesn't expose.
Eduardo
training, I would say that triple n-back is miles better than dual n-
back.
I can easily go up to 7 n-back without using grouping strategies (I
just 'see' the last n squares in my mind screen and constantly update
them in my mental grid), but when doing triple n-back I can't go past
3-Back. And when I'm finished, I can feel that my brain was taxed like
hell from the triple n-back task.
Shamanu999
The color part could be limiting the faster progress of the other n-back tasks.
What is the highest level one got with triple n-back?
Eduardo schrieb:
Denis Gorodetskiy
but add geometrical shape or 3d model as fourth axis to n-back task
along with color, sound and position.
It could tighten the memory exercise making it more effective.
Shape is kinda natural object to memorize because brain likes physical
things and doesn't like abstractions, isn't it?
Eduardo
:)
Vlad
too. I configured BW to do sessions of triple n-back of 150 trials
each (which means in 30 minutes you do 3 sessions), level manually set
on n=4. Not training regurally (maybe 10 days in a month), I'm now
doing this triple-4-back at 50%, and progressing slow but steady
(something like 2 percent per session - Eduardo you're too good ;). I
like it more than dual-back, but our ultimate goal is IQ gain, am i
right? :) And in one research (Unsworth, Engle, 2008) I read that WM
property that correlates the most with ravens tests is accuracy, not
the number of switches to different stimuli, or the speed of
switching. My guess is, that there should be some optimum between
widening your attention bottleneck (different stimuli) and deepening
your WM (higher n-level). So hexa-n-back would be too much :) but
triple is probably better than dual.
Btw my attempt to replicate the original study is in its final phase,
after 19 days of training people are doing usually 2-3 points better
on complete RAPM II (which in that specific cases looks comparable
with original study). Lets see control group results, what amount of
this is just retest effect. I'll post more in few days.
Eduardo
was below 50%. It is the hardest game I have ever played. I think that
I'll continue to do it until I feel my working memory quality is
enough that I can remember the colors as well as I can maintain in my
mind position and sound.
How many people are in your study Vlad? Maybe we could try doing the
same study with triple n-back vs dual n-back.
Gwern Branwen
>
> exactly, and I agree with triple n-back being definitely more tiring
> too. I configured BW to do sessions of triple n-back of 150 trials
> each (which means in 30 minutes you do 3 sessions), level manually set
> on n=4. Not training regurally (maybe 10 days in a month), I'm now
> doing this triple-4-back at 50%, and progressing slow but steady
> (something like 2 percent per session - Eduardo you're too good ;). I
> like it more than dual-back, but our ultimate goal is IQ gain, am i
> right? :) And in one research (Unsworth, Engle, 2008) I read that WM
> property that correlates the most with ravens tests is accuracy, not
> the number of switches to different stimuli, or the speed of
> switching. My guess is, that there should be some optimum between
> widening your attention bottleneck (different stimuli) and deepening
> your WM (higher n-level). So hexa-n-back would be too much :) but
> triple is probably better than dual.
Unsworth being http://www.psychology.gatech.edu/renglelab/Unsworth%20and%20Engle%20JEPLMC%202008%20Garavan%20paper.pdf
?
--
gwern
Gwern Branwen
> property that correlates the most with ravens tests is accuracy, not
> the number of switches to different stimuli, or the speed of
> switching. My guess is, that there should be some optimum between
> widening your attention bottleneck (different stimuli) and deepening
> your WM (higher n-level). So hexa-n-back would be too much :) but
> triple is probably better than dual.
OK, I've read through Unsworth (and uploaded as usual). At first, I
thought this might indicate that shortening the time for each round
was useless: if there was no relation between reaction speed and
performance, then one would expect lowering allowed reaction time to
have no impact on performance, or negative impact.
But I played some more and thought about it some more, and thought
about Unsworth's point that it was accuracy that make WM useful and
distinguished between high & low performers. I think the value of
shortening time is that it forces you to get it right the first time;
you can't retrieve it, think 'that's not right', and rehearse the
sequence again until you're sure you are right. You don't have enough
time. You retrieve it once, answer, and you're either right or wrong.
Further, it forces focus to an even greater extent. Truly, you can't
drift away during n-back, but 1.5s rounds punish even very brief
drifts. The two factors are, I think, valuable.
(And of course, sessions taking less time is a good encourager to do
them at all! I play n-back much more reliably now that I'm only
sinking in 1 minute or so per session, and can do my daily quota of 5
sessions in under 10 minutes.)
Pushing down the time has held me back at 3-back for a long time now,
but I'm finally hitting 90-100% scores (in the initial session, at
least - I still fall down to ~50 in the last few sessions); the funny
thing is, I notice that I am often remembering the 4th back pair, even
though I have no use for it in 3-back. So I predict I'll start off
with fairly high scores in 4-back, with the slow default rounds.
--
gwern
Pheonoxia
intervals. When I first started, each time I literally recited the
pattern in my head during those three seconds, which allowed me to
advance. Once you get to a high enough level, this becomes
impractical, but I believe the function/strategy becomes subconscious
instead of intentional. One needs to give this function time to
develop into a subconscious effort. Somebody argue against this if you
think I'm wrong.
Gwern, it somewhat bothers me that an avid poster and highly
contributing member to this group does so little n-back. You only do
five sessions per day? That's shitty. I did 30 46-second sessions of
TNB yesterday alone, which amounts to 23 minutes of solid n-backing.
Yes, it's boring and arduous, but just force yourself to do 10 at a
time and conjuring up such willpower in the future won't be so
difficult. Just keep hitting space bar after each round until you get
to 10, then take a break and do another 10.
Curtis Warren
I'd just like to chime in and point out that procrastination does not
work that way. Your advice boils down to "try harder," which, believe
me, as profound advice as it is, has already. been. tried. several
times over. Take me, for example. I had to write a program to _force_
myself to train daily. If "try harder" actually worked, I would've
preferred to do that over spending a day wrestling with the win32 API
process management functions.
Sorry, pet peeve. As you were.
Pheonoxia
should do more sessions, not try harder doing the same amount of
sessions.
Gwern Branwen
> I wouldn't recommend lower level n-backers decrease the time
> intervals. When I first started, each time I literally recited the
> pattern in my head during those three seconds, which allowed me to
> advance. Once you get to a high enough level, this becomes
> impractical, but I believe the function/strategy becomes subconscious
> instead of intentional. One needs to give this function time to
> develop into a subconscious effort. Somebody argue against this if you
> think I'm wrong.
Mm. I don't know how we could decide either way; I could also easily
argue that smaller intervals is good for beginners because besides
encouraging them by making N-back less of a time commitment, it makes
it more difficult for them to accidentally remember back too many
levels. (I remember that on 2 or 3, many errors were because I was
reacting to a stimulus that was identical with another one, but not
far enough back, or too far back.)
> Gwern, it somewhat bothers me that an avid poster and highly
> contributing member to this group does so little n-back.
What can I say? The studies and discussion are more interesting than
the actual n-backing. :)
> You only do
> five sessions per day? That's shitty. I did 30 46-second sessions of
> TNB yesterday alone, which amounts to 23 minutes of solid n-backing.
> Yes, it's boring and arduous, but just force yourself to do 10 at a
> time and conjuring up such willpower in the future won't be so
> difficult. Just keep hitting space bar after each round until you get
> to 10, then take a break and do another 10.
I aspire to 10, actually, but 5 is my minimum. Part of the problem is
that I've been trying to do it before going to bed, but I go to bed
when sleepy... You see the issue!
But it's not quite that bad. As I think I mentioned once before, I
doubled the trials to 40+3; so 5 is in effect 10. (So I'm doing a
third what you did yesterday; if it takes about 7 minutes per my 5,
3*7 is 21, or ~23 minutes.)
--
gwern
Pheonoxia
study you posted says that those who practiced in the morning made 40%
gains while those before bedtime made 48% gains. While it is better to
train later, training early provides 83.33% of the same benefit, which
to me is mostly insignificant, so if I have the time early in the day,
I train early. It also gives me more motivation to train just before I
go to bed.
You say you're hearing impaired in both ears? So is my older brother,
but I can't get his dumbass to try n-back for the life of him. Are you
able to distinguish the aural stimuli as unique from one another? He
thinks he's so fucking smart, that his only faults are his two ears,
so when I suggest he do something that is scientifically proven to
make him smarter his ego and arrogance get in the way.
Gwern Branwen
> Well that's good. I also try to n-back before bedtime, but that sleep
> study you posted says that those who practiced in the morning made 40%
> gains while those before bedtime made 48% gains. While it is better to
> train later, training early provides 83.33% of the same benefit, which
> to me is mostly insignificant, so if I have the time early in the day,
> I train early. It also gives me more motivation to train just before I
> go to bed.
Yeah. I've wondered if the difference was significant enough to
bother. It may be for me that doing it before bed is just a bad time.
> You say you're hearing impaired in both ears? So is my older brother,
> but I can't get his dumbass to try n-back for the life of him. Are you
> able to distinguish the aural stimuli as unique from one another?
I can, but with a little difficulty. For example, the 'o' and 'l'
pronunciations* sound almost identical to me. This isn't *too* big a
deal; certainly it's not a barrier to doing n-back. (I'd guesstimate
that it occasionally costs me 1 or 2 omissions/errors.)
> He
> thinks he's so fucking smart, that his only faults are his two ears,
> so when I suggest he do something that is scientifically proven to
> make him smarter his ego and arrogance get in the way.
Unfortunately, one of the curses of hearing impairment is that it
affects the brain (I'm assuming here that the hearing impairment was
congenital; I don't know whether this applies to those impaired later
in life).
Even with a perfectly good, adjusted pair of hearing aids (so one
might think the hearing impairment immaterial) I still can't do the
'cocktail party effect' - rather unfortunate, since many social
occasions are in noisy circumstances where one absolutely need to be
able to do that. So if he really thinks it's just his ears that are
affected, he's mistaken.
* At least, I think that's what they are. Exactly what the pair of
sounds is that confuses me is still something of a mystery.
--
gwern
Chris
find that my average score for the first 5 days of triple n-back was
exactly two thirds of my average score for the last 5 days of dual n-
back!