Transcript slides

Cross-task Prediction of Working Memory Performance:
Working Memory Capacity as Source Activation
Larry Z. Daily, Marsha C. Lovett, and Lynne M. Reder
Carnegie Mellon University
Thanks to Scott Filipino for assistance in collecting the data.
This work supported grant F49620-97-1-0455 from the Air Force Office of Scientific Research
and grant N00014-95-1-0223 from the Office of Naval Research.
Working Memory in ACT-R
• Provides the resources needed to retrieve and maintain
information during cognitive processing (Baddeley, 1986)
• Working memory capacity is limited
– W is limited (Anderson, Reder, & Lebiere, 1996)
• Working memory limits vary across individuals
– W varies over individuals (Lovett, Reder, & Lebiere, 1999)
Prior Work:
The MODS Task
• MOdified Digit Span
a
• Read strings of digits
and numbers aloud
• Remember the
numbers
j
2
1st
string
T
b
i
I
M
E
e
6
2nd
string
c
f
8
• Memory set size
varied from 3 to 6
3rd
string
recall
_ _
_
MODS Task
Aggregate Results
MODS Task
Individual Accuracy
Subject 221
W = 0.7
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Subject 201
W = 0.9
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
3
4
5
6
Memory Set Size
Subject 211
W = 1.0
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
3
4
5
6
Memory Set Size
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
3
4
5
6
Memory Set Size
Data
Model
Subject 203
W = 1.1
3
4
5
6
Memory Set Size
MODS Task
Individual Serial Position
CAM Inductive
Reasoning Subtest
Cross-task Correlations
• Estimates of W strongly correlated with
scores on CAM inductive reasoning subtest
– r = 0.50
r2 = .25
• Correlations are a weak test of W’s
predictive ability
• No model of the CAM
New Work:
The n-back task
• Subjects view a long sequence of letters
• For each, indicate whether it is a target or
non-target
• Targets defined by condition
– 0-back - target is a letter given at the start
– 1-back - letter is a target if it matches the previous letter
– 2-back - letter is a target if it matches the letter before
the previous one
– 3-back - letter is a target if it matches the letter 3 before
the current one
N-back Strategies
• Familiarity-based
– Subjects use familiarity to decide whether an
item is a target
– Doesn’t depend on W
• Update
– Subjects actively try to maintain a list of prior
items
– Does depend on W
– We model this
The n-back Model
• Encodes the item currently in vision
– Memory chunk encodes item and position
• Attempts to match the item to a stored
memory
– Looks for a memory chunk with a position that
matches the n-back condition and the correct
item
– Sets a flag to indicate match or not
The n-back Model
• Responds
– Target or non-target based on flag
• Updates memory chunks
– Changes the position of each chunk up through
the current n-back condition
– If this process fails, several responses required
to get back on track
Aggregate Results
MODS Task
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
n-back Task
1.00
0.95
0.90
0.85
0.80
0.75
0.70
0.65
0.60
0.55
0.50
Data
Model
3
4
5
Memory Set Size
6
Data
Model
0
1
2
Memory Load
(n-back)
3
Individual Results:
MODS Task
Subject 610
W = 0.8
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Subject 619
W = 0.9
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
3
4
5
6
Memory Set Size
Subject 613
W = 1.0
Subject 623
W = 1.1
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
3
4
5
6
Memory Set Size
3
4
5
6
Memory Set Size
Data
Model
3
4
5
6
Memory Set Size
Individual Results:
MODS Task
Individual Results:
N-back Task
Subject 610
W = 0.8
1.00
0.95
0.90
0.85
0.80
0.75
0.70
0.65
0.60
0.55
0.50
Subject 619
W = 0.9
1.00
0.95
0.90
0.85
0.80
0.75
0.70
0.65
0.60
0.55
0.50
0
1
2
3
Memory Load
(n-back)
Subject 613
W = 1.0
1.00
0.95
0.90
0.85
0.80
0.75
0.70
0.65
0.60
0.55
0.50
0
1
2
3
Memory Load
(n-back)
1.00
0.95
0.90
0.85
0.80
0.75
0.70
0.65
0.60
0.55
0.50
0
Data
Model
Subject 623
W = 1.1
1
2
3
Memory Load
(n-back)
0
1
2
3
Memory Load
(n-back)
Individual Results:
N-back Task
Subject 610
W = 0.8
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Subject 613
W = 1.0
Subject 619
W = 0.9
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0
1
2
3
Memory Load
(n-back)
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0
Hits
Subject 623
W = 1.1
1
2
3
Memory Load
(n-back)
False Alarms
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0
1
2
3
Memory Load
(n-back)
Predicted Hits
0
1
2
3
Memory Load
(n-back)
Predicted False Alarms
Conclusions
• Varying W captures differences in
individuals’ working memory performance.
• W can be used to predict performance
across tasks.
• W is a workable measure of working
memory capacity.
• ACT-R can be fruitfully applied to the study
of individual differences.