"We agree that models, like data, should be freely available according to the normal standards of science, but caution against confusing implementations with specifications. Models may embody theories, but they generally also include implementation assumptions.' 2/n
-
-
Prikaži ovu nit
-
"Cognitive modeling methodology needs to be sensitive to this. We argue that specification, replication and experimentation are methodological approaches that can address this issue." 3/npic.twitter.com/GSqcEpin8F
Prikaži ovu nit -
"Computational modeling has been a central method in cognitive science since Newell, Shaw and Simon’s seminal work on human problem solving (Newell, Shaw, & Simon, 1958). However, a recent debate initiated by McClelland (2009) on the place of computational modeling ... " 4/n
Prikaži ovu nit -
"...has brought into focus some limitations of the modeling enterprise as currently practiced. Perhaps most critically, one subtext of McClelland’s discussion is that many aspects of modeling are not well understood by those who are not trained in its methods." 5/n
Prikaži ovu nit -
"modeling has more to offer cognitive science than just the investigation of the implications of complex ideas. A central aim of cognitive science is the development of theories of the computational processes underpinning intelligent behavior, ... human, animal or artificial" 6/n
Prikaži ovu nit -
"Comp methods have the potential to play a central role in the specification of such theories, as the development of implementations of theories allows the simulation of the behaviors of interest (over & above the exploration of implications of sets of ideas or assumptions)." 7/n
Prikaži ovu nit -
"In some cases it may be appropriate to equate a model (i.e., a specific implementation) with a theory (....). However, in general, implementations are not theory specification." 8/n
Prikaži ovu nit -
"At worst the prescriptions [that code is made available] could undermine modeling by promoting a culture outside of the modeling community of accepting implementations at face value" 9/n (IMO a really important point, and may apply also to some of the current DL discussions)pic.twitter.com/jGwZuykRRp
Prikaži ovu nit -
"(...) as McCloskey (1991) notes, “the characterization of [a model] should provide a basis for assessing the extent to which successes or failures in simulating particular phenomena reflect theory-relevant or theory-irrelevant features [of the model]” (p. 389)." 10/n
Prikaži ovu nit -
"Providing computationally naive researchers an implementation without also providing them with the means to discriminate between the consequences of theory-relevant or -irrelevant assumptions raises possibility that such researchers will overinterpret the model’s behavior" 11/n
Prikaži ovu nit -
"This can work in 2 ways: being excessively impressed with some positive aspect of the model’s behavior that is not due to the theoretically critical elements of the model, or unduly critical of some negative aspect that is a consequence of a minor implementation decision." 12/n
Prikaži ovu nit -
(I feel these are vital points. Not only for modeling novices, but also modeling experts. Speaking for myself, it's really difficult to strike right balance such that readers take away exactly what's supported by formal analysis, not more nor less) 13/n
Prikaži ovu nit -
"...it is highly plausible that an easy-to-use graphical user interface will only exacerbate the problem of distinguishing theoretical consequences from the implications of a specific implementation, as computationally naive users are unlikely to go beyond the interface." 14/n
Prikaži ovu nit -
"The more critical issue is that implementations must be computationally complete, in the sense that they must be specified to a level that allows the implementation to be run. Cognitive-level theories do not specify all aspects of an implementation ..." 15/n
Prikaži ovu nit -
"(...) so cognitive models, as implementations, include implementation details – details of the implementation that are necessary for the implementation to be run but that are not part of the cognitive-level theory." 16/n
Prikaži ovu nit -
(Honestly, I think even modellers often don't know exactly which parts of their implementations are core theoretical commitments & which ones not. Again, speaking for myself: sometimes I discover commitments I made without realizing, that turn out core or turn out irrelevant) 17/pic.twitter.com/0Hg9kdujW2
Prikaži ovu nit -
(But finding out whether or not the irrelevant ones do not matter can be a core theoretical exercise. That way we can discover how some of our model properties may be invariant. Such model invariants may have particular explanatory and predictive power.) 18/n
Prikaži ovu nit -
"Models must therefore be distinguished from specifica- tions of theory. How can computational modeling assist in the promotion of this distinction and in the clarification of cognitive theory?" 19/n
Prikaži ovu nit -
"We propose three methodological approaches that modeling might adopt to achieve this goal: abstract specification, reimplementation, and sensitivity analyses via computational experimentation." 20/n
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.
