Presentation is loading. Please wait.

Presentation is loading. Please wait.

Dynamic Decision Making Laboratory Carnegie Mellon University 1 Social and Decision Sciences Department ACT-R models of training Cleotilde Gonzalez and.

Similar presentations


Presentation on theme: "Dynamic Decision Making Laboratory Carnegie Mellon University 1 Social and Decision Sciences Department ACT-R models of training Cleotilde Gonzalez and."— Presentation transcript:

1 Dynamic Decision Making Laboratory Carnegie Mellon University 1 Social and Decision Sciences Department ACT-R models of training Cleotilde Gonzalez and Wai-Tat Fu Dynamic Decision Making Laboratory (DDMLab) www.cmu.edu/ddmlab Carnegie Mellon University

2 Dynamic Decision Making Laboratory Carnegie Mellon University 2 Social and Decision Sciences Department Agenda Brief intro to ACT-R and the goals of MURI project Communication with experiments –Data collection and modeling –Design of experiments using CMU Radar simulation Modeling work: data fitting –Healy, Kole, Buck-Gengler and Bourne, 2004: Prolonged work will result in distinctive effects on RT and performance (speed-accuracy tradeoff). –Buck-Gengler & Healy, 2001 –Kole, Healy, Buck-Gengler 2005 Modeling work: predictions –Duration –Depth of processing –Repetition

3 Dynamic Decision Making Laboratory Carnegie Mellon University 3 Social and Decision Sciences Department ACT-R models of training: Goals Determine the cognitive functions and mechanisms corresponding to training principles Create computational models that will be used as predictive tools for the effect of training manipulations Develop an easy-to-use graphical user interface to help: –manipulate a set of training and task parameters –determine speed and accuracy as a result of the parameter setting and the training principles

4 Dynamic Decision Making Laboratory Carnegie Mellon University 4 Social and Decision Sciences Department The 2x2 levels of ACT-R http://act.psy.cmu.edu (Anderson & Lebiere, 1998)

5 Dynamic Decision Making Laboratory Carnegie Mellon University 5 Social and Decision Sciences Department Space processes

6 Dynamic Decision Making Laboratory Carnegie Mellon University 6 Social and Decision Sciences Department Representation and Equations RetrievalGoal ManualVisual Productions IntentionsMemory Motor Vision World Activation Learning Latency Utility Learning IF the goal is to categorize new stimulus and visual holds stimulus info S, F, T THEN start retrieval of chunk S, F, T and start manual mouse movement S201 Size Fuel Turb Dec L203Y Stimulus Chunk BiBi S SL S 13

7 Dynamic Decision Making Laboratory Carnegie Mellon University 7 Social and Decision Sciences Department Chunk Activation base activation =+ Activation makes chunks available to the degree that past experiences indicate that they will be useful at the particular moment: Base-level: general past usefulness Associative Activation: relevance to the general context Matching Penalty: relevance to the specific match required Noise: stochastic is useful to avoid getting stuck in local minima associative strength source activation ( ) * similarity value mismatch penalty ( ) * ++ noise

8 Dynamic Decision Making Laboratory Carnegie Mellon University 8 Social and Decision Sciences Department Data collection and modeling

9 Dynamic Decision Making Laboratory Carnegie Mellon University 9 Social and Decision Sciences Department Model 1: cognitive model of data entry From Healy, Kole, Buck-Gengler and Bourne, 2004: –Prolonged work will result in distinctive effects on RT and performance (speed-accuracy tradeoff).

10 Dynamic Decision Making Laboratory Carnegie Mellon University 10 Social and Decision Sciences Department Speedup: Production compilation + faster access to key loc Accuracy ↓: ↑ Noise in retrieval mechanism Encode next number Retrieve key location Type next number Hit Enter All numbers encoded All numbers typed Initiation time Execution time Conclusion time Wrong digits: Noise in retrieval Extra/Missed digits: Noise in stopping criteria Model 1: How does it work?

11 Dynamic Decision Making Laboratory Carnegie Mellon University 11 Social and Decision Sciences Department Production compilation Basic idea: –Productions are combined to form a macro production  faster execution –Rule learning: Retrievals may be eliminated in the process Practically: declarative  procedural transition Data Entry task: –From: Visual  Retrieval (key loc)  Motor –To: Visual  Motor

12 Dynamic Decision Making Laboratory Carnegie Mellon University 12 Social and Decision Sciences Department Model fit Proportion correct Total RT

13 Dynamic Decision Making Laboratory Carnegie Mellon University 13 Social and Decision Sciences Department Conclusions from Model 1 Production compilation results in faster execution Fatigue may correspond to the increase of activation noise (in retrieval and stopping), producing more errors and a decrease in the proportion of correct responses

14 Dynamic Decision Making Laboratory Carnegie Mellon University 14 Social and Decision Sciences Department Model 2: Long-term repetition priming From Buck-Gengler and Healy, 2001: –Depth of processing: Old numbers typed faster than new numbers Trained using word format  faster at testing than those trained using numbers –Word format  more elaborate processing  better retention of skills –Effect seems to depend on “type of processing”, not just amount of encoding time

15 Dynamic Decision Making Laboratory Carnegie Mellon University 15 Social and Decision Sciences Department Encode 4 digits Buck-Gengler and Healy, 2001, methods Type digitsHit Enter Training stage Testing stage 1 week delay Old Numbers New Numbers Word Numerals Word Numerals X W or w/o Articulatory Suppression Number row Keypad (motor process) Response: Numbers Letters (cognitive) Retention of skills depends on: 1.Cognitive or Motoric processes? 2.Depth of processing? 3.Mental fatigue?

16 Dynamic Decision Making Laboratory Carnegie Mellon University 16 Social and Decision Sciences Department Encode next number Retrieve key location Type next number Hit Enter All numbers encoded All numbers typed Initiation time Execution time Conclusion time Retrieval process in ACT-R Model 2: how does it work?

17 Dynamic Decision Making Laboratory Carnegie Mellon University 17 Social and Decision Sciences Department “THREE” “3” Episodic representation of stimuli THREE (Semantic Concept) L3 (Key location) “T” (Key location) Semantic processing leads to deeper processing of stimuli Deep processing

18 Dynamic Decision Making Laboratory Carnegie Mellon University 18 Social and Decision Sciences Department Semantic Priming “THREE”“3” THREE (Semantic Concept) or L3 retrieve Semantic concept boosting retrieval of key location Faster retrieval when trained in word format S ji

19 Dynamic Decision Making Laboratory Carnegie Mellon University 19 Social and Decision Sciences Department RT by block, response and presentation formats (at training)

20 Dynamic Decision Making Laboratory Carnegie Mellon University 20 Social and Decision Sciences Department Presentation format continuity (at test)

21 Dynamic Decision Making Laboratory Carnegie Mellon University 21 Social and Decision Sciences Department Conclusions from Model 2 The long-term repetition priming (RP) can be explained by semantic priming mechanism in ACT-R The results from the word and letter conditions suggest that it is not the amount of processing, but the “type” of processing that leads to the RP

22 Dynamic Decision Making Laboratory Carnegie Mellon University 22 Social and Decision Sciences Department Model 3: Suppression of vocal rehearsal From Kole, Healy, Buck- Gengler 2005 –less encoding (shallow processing) Old numbers still faster than new numbers –But: Word format at training  not significantly faster Deep processing  better retention and durability of skills

23 Dynamic Decision Making Laboratory Carnegie Mellon University 23 Social and Decision Sciences Department Kole et al.

24 Dynamic Decision Making Laboratory Carnegie Mellon University 24 Social and Decision Sciences Department Model predictions: three factors Delay –How performance deteriorates with time after training Depth –How different depth of processing (training) may affect the retention of skills Repetition –How re-training may help retention of skills

25 Dynamic Decision Making Laboratory Carnegie Mellon University 25 Social and Decision Sciences Department Prediction 1 - Delay

26 Dynamic Decision Making Laboratory Carnegie Mellon University 26 Social and Decision Sciences Department Prediction 2 – depth of processing

27 Dynamic Decision Making Laboratory Carnegie Mellon University 27 Social and Decision Sciences Department Prediction 3 - Repetition

28 Dynamic Decision Making Laboratory Carnegie Mellon University 28 Social and Decision Sciences Department Conclusions of predictions Training effects decay approx exponentially with time More extensive training leads to better retention of skills Re-training may be more efficient that extensive initial training for retention of skills

29 Dynamic Decision Making Laboratory Carnegie Mellon University 29 Social and Decision Sciences Department Next steps Integrate a set of model parameters for the data entry task and ACT-R parameters into a prediction tool Continue participating in the development of new experiments using the Radar task Integrate ACT-R predictions to IMPRINT


Download ppt "Dynamic Decision Making Laboratory Carnegie Mellon University 1 Social and Decision Sciences Department ACT-R models of training Cleotilde Gonzalez and."

Similar presentations


Ads by Google