week 8 - cognitive models of visual working memory

Question 1

What do function/mathematical term do LSE and MLE minimise?

Answer 1

LSE minimises the RMSD
MLE minimises a reversed log-likelihood function

Question 2

Which model fit addresses some of the issues in LSE?

Answer 2

MLE -> can meaningfully compare multiple/different model fits

Question 3

What is visual working memory (VWM)?
What is its scientific definition?

Answer 3

visual information processing in the current moment (now)
active maintenance of visual information to serve the needs of ongoing tasks

Question 4

What paradigm is used with the DSM to test VMW?

Answer 4

Change Detection Paradigm (CDP): show image then a short break. then show altered image and participant asked if the two images are the same or different?

Question 5

What is the main focus of VWM research?
What is the average capacity in the VWM?

Answer 5

why is the vwm so limited?
average 4 items held in the VWM

Question 6

Why is there a big interest in researching the capacity of th VWM?

Answer 6

VWM capacity is predictive of how well they perform in other skills in fluid attention, problem solving, reading ability, text comprehension

Question 7

What does the Discrete-slots Model (DSM) try to explain about the capacity of the VWM?

Answer 7

-VWM capacity limit arises from a limited resource.
-limited resource is allocated to a LIMITED number of DISCRETE representation slots
-no information is stored about additional items once the capacity limit is reached

Question 8

What is the DSM in layman’s?

Who created this model and when?

Answer 8

-you have a limited number of discrete slots to store information in your VWM in your brain
(one item of info per slot = discrete slots)
-when using VWM, you store information in each slots until your capacity limit is reached (even if there are still empty slots)

Zhang and Luck 2008

Question 9

Is the DSM an all-or-nothing type idea?

Answer 9

yes, either you remember all of one item or nothing (can’t remember half an item)

Question 10

Who created the Continuous-resource model (CRM) of VWM?
How does it model VWM capcity?

Answer 10

Bays et al. 2009
equal distribution of a continuous resource(energy to remember item) across all items visually presented
but there is a capacity limit so items lose precision

Question 11

What happens to the distribution of continuous resource as the number of items increases, in the CRM?
Why?

Answer 11

-as no. of items increases, representations (store of an item) loses resource/remembered worse
-because all items are allocated the same amount of resource and there is a capacity limit

Question 12

What is the difference between the Discrete-slots model and the Continuous-resource model?

Answer 12

all-or-nothing way of storing items in DSM means that you forget items completely once capacity is reached but CRM always remembers at least a part of every item due to equal resource allocation

Question 13

Which paradigm fixes some of the issues with the Change Detection Paradigm (CDP)?
Why is it better?

Answer 13

-Continuous Reproduction Paradigm: uses a colour wheel in which participants can select which colour they remembered from the initial image
-you can quantitively measure precisely how far off the participant’s response was from actual/initial colour -> the number computed can say whether participant remembered nothing of the item (all-or-nothing) or part of item (continuous) (can model CRM and DSM)

Question 14

What are the assumptions of mixture models (use example for VWM)?

Answer 14

assume you have more than one processes going on eg. no. of slots and precision of representation AND that these different processes have different distributions

also can replace processes for populations

Question 15

What are the mixture models?

Answer 15

-Two-components mixture model/Standard-mixture model
-three-component mixture model/swap model

Question 16

What are the two components of the standard-mixture model? (use an example the CDP test)
How is each component represented graphically?

Answer 16

-noisy target representation: educated guess (you remember part of the item) -> Von Mises distribution
-random guessing: (choose any colour on wheel) -> uniform distribution (block)

Question 17

For the standard-mixture model, why is the noisy target representation a Von Mises distribution?

Answer 17

because there is a lose fuzzy memory around the target of the peak -> skinny bell shape

Question 18

In the standard-mixture model equation, what does it calculate?
What does the Kappa calculate?
What does pu figure represent/calculate?
What does pt represent/calculate?

Answer 18

-probability of the given response (response from participant)
-dispersion of Von Mises distribution -> how wide curve is
-probability of guessing -> possibility of random guessing which is uniformly distributed (block)
-probability of actually recalling target/actual colour (von Mises dist)

Question 19

As the Kappa changes in the Standard-mixture model, what happens? and in laymans?

Answer 19

Kappa increases, the less precise the internal representation of item (how fuzzy colour is stored in brain

Question 20

What is Kappa?

Answer 20

the precision of the internal representation

Question 21

What are all the parts to a standard-mixture model equation?

Answer 21

probability of given response = probability of recalling the target * target representation + probability of guessing

𝑝(𝜃̂ )=(1 - 𝑝𝑢 ) * ϕκ (𝜃̂ - 𝜃) + 𝑝𝑢 1/2𝜋

Question 22

What did the study in 2013 by Luck and Vogel say about the goodness of model fit for the standard-mixture model?

Answer 22

observations data do follow some trend of the predicted data -> probability to remember something decreases once you reach capacity limit

but not quite the same shape of lines

Question 23

What is the Swap Model aka?
What is the difference between swap model and standard-mixture model?

Answer 23

-three-component mixture model
-builds on standard-mixture model with extra non-target colour distribution (colours not encompassed by von mises)

Question 24

What are the parts of the swap model equation?

Answer 24

probability of given response = probability of recalling target*target representation + probability of random guessing + non-target representation(s)

Question 25

Why is non-target representaion(s) plural ?

Answer 25

as there can more 1+ non-target colours but only one target colour

Question 26

Does Kappa size change in the different parts of the swap model? why?

Answer 26

kappa stays the same between the target and non-target representations as swap model represents CRM and all items are allocated equal resource

Question 27

Swap Model What happens to the precision of the representation of items (in the brain) as the set size (the number of things you have to remember) increases? What does the graph of this relationship look like? Why does this happen?

Answer 27

-set size increases, precision of representation decreases -x = no. of items, y=precision, decreasing curve -because you have to share same limited resource across more items (like CRM)

Question 28

What are the two cognitive modelling approaches? What the aims for each approach?

Answer 28

-explanatory models and measurement models explanatory: to explain processes of cognition that underly observations of behaviour. measurement: to quantify or describe latent cognitive variables based on observed data.

Question 29

What is the difference between how the explanatory and the measurement approach to cognitive modelling fit models? In doing this, what is the aim of each? What sort of parameters does each have?

Answer 29

explanatory fits model simultaneously to multiple experimental conditions because to explain differences between these exp conditions (aim) fixed parameters measurement fits model separately to multiple experimental conditions because to asses how these conditions affect latent constructs majority free parameters

Question 30

What is a latent construct?

Answer 30

an invisible, theoretical variable that is measured indirectly through observable indicators like intelligence, personality traits

Question 31

What is the benefit of using an explanatory modelling approach to mixture models?

Answer 31

you can also investigate WHY these cognitive processes are happening

Question 32

What additional component does the 3c model have which the 2c doesnt have?

Answer 32

pn = non-target distribution, models swapping

Question 33

What is swapping? Why is swapping interesting? Which model represents swapping?

Answer 33

-cognitive process unique to WM where an item's feature are swapped: for example location and colour are a part of an item -> green is associated with top left corner in paradigm but participant swaps green for orange -swapping doesnt happen in STM -3c model: pn = non-target dist for swapping

Question 34

What helps run these mixture models?

Answer 34

pre-existing packages like mixtur whcih runs model simulation and model recovery

Question 35

For CRP, what is the unit for the colour wheel response?

Answer 35

radians or degree of error

Question 36

What are on the axes of 2c and 3c graphs of the models?

Answer 36

x axis = possible colours/ participant responses y axis = response probability

Question 37

3c model: do both the target and non-target representations uses Von Mises distributions?

Answer 37

yes both use von mises

Question 38

AS set size increases from 1-6, what happens to value for each parameter: Kappa, pt, pu and pn?

Answer 38

Kappa= increases pt = decreases pu = increases pn = increases

Question 39

How could you compare the 2c or 3c model to see which has the best model fit? Why this method in particular?

Answer 39

-use the likelihood-ratio test (LRT) - a MLE method -because 2c and 3c are nested models

Question 40

What is parameter recovery?

Answer 40

-fitting a model to synthetic (simulated) data generated with known parameter values in order to test how accurately those parameters can be estimated by the model-fitting procedure. -to test how good is the model is at estimating parameter values? evaluating your real parameter values with simulated parameter values

Question 41

What are the four steps in parameter estimation?

Answer 41

1)choose plausible parameter values 2)simulate synthetic data with these parameters 3)fit model to synthetic data 4)compare chosen parameter values (step1) with parameter values estimated in step 3

Question 42

As the number of trials increases, what happens to the accuracy of parameter recovery?

Answer 42

model gets better at estimating parameter values

Question 43

What can mixtur help do?

Answer 43

-multiple model comparisons to see which one has best fit -parameter recovery by simulating data

Question 44

What was the focus of the study by Jiange et al. 2023? What were the two ways participants enhanced their performance?

Answer 44

-investigated the training effects on VWM representations (how items are stored in brain) -capacity and precision (but are trade-off of each other)

Question 45

Why did Jiange et al 2023 have to use non-nested model comparison? What model comparison test did they use?

Answer 45

because the 2c and 3c model had completely different parameter and parameter values Used Bayesian Information Criteria BIC

Question 46

What do BIC values mean in terms of model fit? is a BIC value given to every model compared?

Answer 46

lower BIC value = better model fit yes every model has its own BIC value calculated

Question 47

Overall for Jiang et al 2023, what was the best model fit, 2c or 3c? what did they discover was interesting about the best model fit for each participant's data?

Answer 47

-2c better -from pre-test to post-test (after training), ~50% participants changed in which model fit their data best (ie from 2c to 3c or vice versa) training changes responses

Question 48

Jiange et al 2023 During training sessions only, what effect did time/more training sessions have on capacity and precision? What is the above question an example of what type of cognitive modelling approach?

Answer 48

time has no effect on capacity, ONLY precision measurement models

Question 49

Jiange 2023 orientation reproduction paradigm, from pre-test to post-test, what happens to precision and capacity? What does this tell you about the effect of training of orientation reproduction paradigm?

Answer 49

precision increases in experimental group but not in active control capacity doesnt change in either groups improvement is driven by changes in precision and not capacity

Question 50

Jiange 2023, what did they discover about strategy used by participants in the orientation reproduction task?

Answer 50

seen in experimental group, increased use of canonical orientations (like 12,3,6,9 oclock) AFTER training maybe this is why they improve their precision

Question 51

What are the conclusion from Jiange 2023?

Answer 51

-training improves precision not capacity -improvements in precision are highly specific to paradigm and stimuli used -changes in response patterns suggest possible interaction between paradigm and stimuli specific expertise

Question 52

How did cognitive modelling help uncover conclusions from Jiange 2023?

Answer 52

-using mixture models of VWM allowed to differentiate between changes in capacity and precision due to training

Question 53

Why is VWM modelled by mixture models?

Answer 53

we assume the VWM is explained by multiple processes -> represented by different parts in mixture models

Question 54

What are the theoretical assumptions of the 2c and 3c model? what theory does it model However which paradigm is used in modelling of VWM?

Answer 54

2c = DSM: items stored as discrete slots (all-or-nothing) 3c = CRM: equal distribution of a continuous resource across all items CRP