Transcript Document
Predicting trained task performance: The interaction of taxonomy, data, and modeling
The function of taxonomic analyses
•
Goal:
Predict and optimize performance on trained tasks
•
Four dimensions of analysis:
1.
Task type
2.
Training methods
3.
Performance measures
4.
Training principles
•
Develop analyses for dimensions that can:
- Relate similar
tasks
- Cover
task
and
training
domains - Capture meaningful aspects of
performance
- Provide useful
generalizations
for optimization constraints
Iterative development process
Analysis framework Task analyses Task models
Target phenomena (including general principles) Empirical training and performance data
Modeling performance using task features:
An earlier approach
Roth, Thomas J. 1992. Reliability and validity assessment of a taxonomy for predicting relative stressor effects on human performance. Micro Analysis & Design Technical Report 5060-1.
• Roth described military tasks as weighted vectors of five features: Attention Perception Psychomotor Physical Cognitive • Task features were found to be non-independent.
• Need task features that are (more) independent and provide more detail about cognition.
A decomposition for cognitive tasks
Perception/attentional processing
Vision, hearing, tactile sensation
Cognitive/affective processing Synthesis
Executive control/Monitoring Memory/Representation Reasoning/Problem solving Motivation/Affect Concept formation Imagery
Response planning
Speech planning Motor planning
Physical/communicative response
Language/Speech Manipulation/Fine motor Action/Gross motor
Modeling the data entry task
•
Task
of data entry is simple, but decomposable:
Perceptual processing
Read number
Cognitive processing
Encode number Plan motor output
Synthesis Planning Response
Type number •
Training
consists of
practice
and
repetition
(1 pass/item) •
Performance
is measured by
speed
and
accuracy
of typing
Understanding componential performance
Consider effects of training
practice
and
repetition
on: Reading Encoding Planning Typing …as measured by
speed
and
accuracy
of data entry.
Empirical phenomena
Fixed processing of perception and response
Reading and typing numbers requires a fixed amount of time, which does not improve with practice.
Encoding improvement of only some percepts
Repeated encoding of numerals does not improve with practice, but (non-usual) encoding numbers from words does.
Repetition priming of motor planning
Repeatedly planning the motor response for numbers leads to specific learning and speeding of responses.
Speed –accuracy trade-off
Increased response speed is associated, for most people, with a decrease in the accuracy of responses.
Modeling implementation of phenomena
• Reading and typing speeds are constant, but depend on format.
t read (n) = c ReadFormat ; t type (n) = c TypeFormat
• Repetitious motor planning speeds responses (according to the power law of learning) and lowers accuracy.
t a planning planning (n) = a p (n) = f(t + b p (N(n,t) + p planning (n)) p ) -
p
• Repetitious encoding only affects speed for less familiar percepts
t encoding (n) = a e + b e (N(n,t) + p e ) -
e a encoding (n) =
?
OR c
Encoding
The IMPRINT simulation
Data Entry
Repeated motor planning Encoding Reading Typing Practice = N
Decomposition
Iteration:
Modeling
Experimentation
•
Fatigue Fact
: Speed of cognitive processing decreases without repetition priming.
Update model to reflect cognitive fatigue from practice.
•
Feedback Fact
: Accuracy can be improved with feedback.
Update model to decouple speed and accuracy functions.
Include monitoring function in task decomposition.
•
Error types Prediction
: Accuracy decline is due to motor planning errors.
Examine effects on accuracy of encoding and planning.
Representations and error types in task may be different.
Iteration:
Enhancing the taxonomic analyses
•
Individual differences Fact
: Not everyone exhibits the speed –accuracy trade-off.
Fact
: Higher cognitive ability leads to faster skill acquisition.
Add a dimension of individual variation.