Comp. Models of the Mind II b Flashcards
Explain Bonini’s paradox!
as a model of a complex system becomes more complete, it becomes less understandable (as hard to understand as real world system)
What do we want to ask ourselves, when we validate a model?
How adequately does the model reflect the aspects of the real world it has been designed to model?
Six factors of the multidimensional utility criterion?
- parsimony
- effectiveness (explicit procedures for deriving predictions)
- broad generality (models based on general cognitive theories also reduce the irrelevant specification problem)
- accuracy and ease of falsification
- surprise! (interesting and counterintuitive behavior)
- coverage of variety of data and different knowledge
Three actions to show how adequately a model reflects the aspects of the real world:
- explicate how much a model constrains the data to befitted
- report data variability: verify real world data agrees also with outcomes ruled out by the model
- show there are plausible results the model cannot fit
Give an example of process analysis!
Marr’s Levels of Explanations of Complex Systems
Marr’s analysis has ____________ three levels.
Marr’s analysis has AT LEAST three levels.
Steps in cognitive modeling according to Busemeyer & Diederich (2010) - step 1/6
REFORMULATE assumptions of conceptual theoretical framework into more rigorous mathematical/computer language form
Steps in cognitive modeling according to Busemeyer & Diederich (2010) - step 2/6
Additional detailed AD HOC ASSUMTPIONS to COMPLETE the model: required for precise quantitative predictions
(e.g. selection of feature definitions)
Steps in cognitive modeling according to Busemeyer & Diederich (2010) - step 3/6
PARAMETER ESTIMATION from observed data
e.g. weight coefficient
Steps in cognitive modeling according to Busemeyer & Diederich (2010) - step 4/6
COMPARISON of predictions of competing models
Steps in cognitive modeling according to Busemeyer & Diederich (2010) - step 5/6
EMPIRICAL TESTS, aiming for parameter-free tests
Steps in cognitive modeling according to Busemeyer & Diederich (2010) - step 6/6
REFORMULATE THEORETICAL framework and construct new models
What is there to say about: Steps in cognitive modeling according to Busemeyer & Diederich (2010) - step 1/6
- Use of basic cognitive principles of the conceptual theory for model construction
What is there to say about: Steps in cognitive modeling according to Busemeyer & Diederich (2010) - step 2/6
- Number of ad hoc assumptions should be minimised
What is there to say about: Steps in cognitive modeling according to Busemeyer & Diederich (2010) - step 3/6
- Ideal: Parameter-free models
What is there to say about: Steps in cognitive modeling according to Busemeyer & Diederich (2010) - step 4/6
- Question whether model CAN fit data is MEANINGLESS!
- Which model provides a better representation wrt. specific aspects of target
What is there to say about: Steps in cognitive modeling according to Busemeyer & Diederich (2010) - step 5/6
- Experimental conditions leading to opposite qualitative or ordinal predictions from competing models for any parameter settings (e.g. different categorization)
- Alternative: quantitative tests: magnitude of prediction errors
Why are Marr’s levels of analysis so important?
Importance of clearly identifying/delineating/distinguishing the DOMAIN of a model
The three levels of Marr:
- Competence / Computational Theory
- Representation and Algorithm
- Hardware Implementation
The aims/questions of the three levels of Marr:
- WHAT is the GOAL of the computation, WHY is it appropriate, and what is the logic of the strategy by which it can be carried out?
- HOW can this computation be implemented? In particular, what is the REPRESENTATION for the input and output and what is the ALGORITHM for the transformation?
- How can the representation and the algorithm be REALISED PHYSICALLY?
The what + why (= computational theory) of the check register:
What: arithmetic, addition (independent of particular representation)
Why: addition meets purposeful constraints (e.g. zero element, commutativity)
The how (representation and algorithm) of the check register:
- Addition: Same representation of numbers for inputs and outputs (or: bar-code -> total sum in numbers)
- Wide choice of representations
- Choice of algorithm often depends on representation
- Quality of algorithm considered if multiple algorithms per representation possible
The physical realisation of the check register:
- (Mis)match of algorithmic styles and computational substrates (e.g. parallelism on single-processor architecture)
Name 7 limitations of GOMS!
- models apply to skilled users only
- only NGOMSL accounts for (restricted) learning
- no account for recall after period of disuse
- no account for slips even skilled users make
- focus on motor components rather than on cognitive processes
- task selection itself is not addressed
- no modelling of fatigue or individual differences
Name three advantages of GOMS!
- prediction of human performance with reasonable accuracy
- widely used for qualitative and quantitative task analysis
- framework for fitting of specialised models