Ravenov test inteligence v kognitivni arhitekturi ACT-R

ANDOLJŠEK, JERNEJ (avtor), Robnik Šikonja, Marko (mentor), Bucik, Valentin (komentor)

Kognitivna arhitektura ACT-R omogoča visokonivojsko simuliranje človeških miselnih procesov pri opravljanju raznovrstnih nalog. V preteklih letih je bila spisana peščica modelov, ki simulirajo delovanje uma med reševanjem Ravenovih matričnih inteligenčnih testov, med drugim tudi v arhitekturi ACT-R. Dosedanji modeli pisani v ACT-R niso zmožni rešiti vseh primerov standardiziranih testov Ravenovih progresivnih matric, tudi kadar odstranijo določene hibe človeškega mišljenja, kot je pozabljanje. V ACT-R smo ustvarili model, temelječ na delu Ragni-ja & sodelavcev (2012), na katerem smo poskušali razumeti težave pri reševanju testov, ter ugotoviti, ali je težavnost reševanja določenih primerov posledica omejitev kognitivne arhitekture ali slabosti modela. Zaključimo, da so zahteve po točnosti reševanja problemov, natančnosti simulacije uma in obvladljivosti kode med seboj v konfliktu.

• Jezik: Slovenski jezik
• Ključne besede: ACT-R, Ravenove progresivne matrice, računalniško modeliranje mišljenja, inteligenčni testi, kognitivne arhitekture
• Vrsta gradiva: Diplomsko delo
• Organizacija: FRI - Fakulteta za računalništvo in informatiko
• Leto izida: 2014
• PID: 20.500.12556/RUL-29971
• Datum objave v RUL: 22.10.2014

Sekundarni jezik

• Jezik: Angleški jezik
• Naslov: Raven's intelligence test in cognitive architecture ACT-R

Izvleček

The ACT-R cognitive architecture allows high level simulations of human mental processes during the execution of various tasks. So far a small number of models exist, that simulate the working of the mind when it is solving Raven's progressive matrices intelligence tests. Some of these models use the ACT-R architecture. The existing ACT-R models are incapable of solving all the standardized Raven's progressive matrices tests, even when removing human limitations such as forgetting. We created a model based on the work of Ragni et al. (2012), with which we tried to understand the problems associated with solving these tests. We try to determine whether the difficulties in solving certain tests stem from the limitations of the cognitive architecture, or are perhaps avoidable with modifications of the model. We find that there is a trade-off between a model's performance at solving tests, its accuracy as a simulation of the mind, and the level of its complexity.

• Ključne besede: ACT-R, Raven's progressive matrices, cognitive modelling, intelligence tests, cognitive architectures