A Computational Model of A Computational Model of Routine - - PowerPoint PPT Presentation
A Computational Model of A Computational Model of Routine Procedural Memory Routine Procedural Memory Frank Tamborello Frank Tamborello Department of Psychology Department of Psychology Rice University Rice University Houston, TX 77005
Frank Tamborello Department of Psychology Rice University Houston, TX 77005 tambo@rice.edu http://chil.rice.edu/
Frank Tamborello Department of Psychology Rice University Houston, TX 77005 tambo@rice.edu http://chil.rice.edu/
Very Brief Introduction Two Experiments, Very Briefly ACT-R Model
2
3
4
SCHEMAS SENSORY- PERCEPTUAL STRUCTURES Sensory Information TRIGGER DATA BASE PSYCHOLOGICAL PROCESSING STRUCTURES External & Internal Actions VERTICAL THREADS HORIZONTAL PROCESSING THREADS Motivational influence on activation Attentional resources add to or decrease activation values
5
GOAL: EDIT-MANUSCRIPT
GOAL: EDIT-SUBTASK
GOAL: ACQUIRE-SUBTASK
✦ GET-NEXT-PAGE
if at end of manuscript page
✦ GET-NEXT-TASK
GOAL: EXECUTE-SUBTASK
✦ GOAL: LOCATE-LINE
– [select: USE-QUOTED-STRING-METHOD – USE-LINEFEED-METHOD]
✦ GOAL: MODIFY-TEXT
– [select: USE-SUBSTITUTE-COMMAND – USE-MODIFY-COMMAND] – VERIFY-EDIT
6
Inputs:
Knowledge
IF-THEN rules (termed “productions”) Declarative knowledge (“chunks”) Subsymbolic parameters
Simulated task environment/world
Output: Time-stamped
7
Task is a routine procedure Subjects trained approximately one week before Concurrent working memory task given
8
1 2 6 ,7 3 4 5
8 9, 11 10 12
1 2 3 , 4 5 6 7
8 9, 11 10 12
static, intervening subtask procedure change, pre-change procedure change, post-change non-intervening semantic control
0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Mean T Mean Total Error Rate
Experiment 1 Condition Experiment 1 Condition
1 2 3 , 4 5 6 7
8 9, 11 10 12
1 2 3 4 5 6 7 8 9, 11 10 12
4 1 3 2 5 6 7 8 9, 11 10 12
static, different-scanner static, same-scanner change procedure, pre-change change procedure, post-change static subtask reordering
0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Mean T Mean Total Error Rate
Experiment 2 Condition Experiment 2 Condition
Model Goal: Simulate error rates across conditions and
4 conditions 14 trial types total not just error generation, but also recovery
Highest human SEM error rate = 0.0415
model should do no worse across the board
16
17
Verify Error Recovery Retrieve Find Move Act Retrieve another action Try again to retrieve the action YES Error? Specify next action NO
18
Verify Error Recovery Retrieve Find Move Act Retrieve another action Try again to retrieve the action YES Error? Specify next action NO Current step = flagged step? NO YES Retrieve New Procedure's Step
static, intervening subtask procedure change, pre-change procedure change, post-change non-intervening semantic control 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Mean Error Rate Experiment 1 Condition humans model
static, different-scanner static, same-scanner change procedure, pre-change change procedure, post-change static subtask reordering 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Mean Error Rate Jammer, Experiment 2 Condition humans model
static, different-scanner static, same-scanner change procedure static subtask reordering 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Mean Error Rate Transporter, Experiment 2 Condition humans model
Discrete, hierarchical goals
governed basic behavior enabled extensible behavior
22
23
Verify Error Recovery Retrieve Find Move Act Retrieve another action Try again to retrieve the action YES Error? Specify next action NO
24
Verify Error Recovery Retrieve Find Move Act Retrieve another action Try again to retrieve the action YES Error? Specify next action NO Current step = flagged step? NO YES Retrieve New Procedure's Step
1, 7 2 6 3 4 5 8 9, 11 10 12
No quantitative, multi-condition error models in
Same model mechanisms across
4 between-subjects conditions 14 trial types
26
Extend model
Step-level error Step completion time
Model training, too
27
Hierarchical, discrete goal representation matters
…for changing circumstances …for error recovery …like CSM
Botvinick & Plaut’s connectionist model too narrow
No postcompletion errors No error recovery No adaptation of old procedures to new circumstances
28
Mike Carissa Chang & Adam Purtee Rick Cooper & Jay McClelland Kristen
29
Questions?
30