Information Technology Reference
In-Depth Information
22%, respectively. The model demonstrated that fewer entities travel from Pho to
M1 directly when the stimuli presented changes from repetitive letters to multidigit
sentences. These results are consistent with the fMRI results of [17].
In practice, because our queuing network model was built with a general structure
with common brain regions, it can be easily transformed to model other task situa-
tions, e.g., PRP [50]. Moreover, the current model can generate behavioral results by
the interaction of the queuing network servers without drawing complex scheduling
charts. These unique features offer great potential of the model for learning and can
easily be used by researchers in cognitive modeling and human factors.
9.3 Modeling the Basic PRP and Practice Effect on PRP with
Queuing Networks and Reinforcement Learning Algorithms
PRP (Psychological Refractory Period) is one of the most basic and simple forms
of dual-task situations and has been studied extensively in the laboratory for half
a century [31]. In the basic PRP paradigm, two stimuli are presented to subjects
in rapid succession and each requires a quick response. Typically, responses to the
first stimulus (Task 1) are unimpaired, but responses to the second stimulus (Task 2)
are slowed by 300 ms or more . In the PRP paradigm of Selst et al. [44], task 1 re-
quired subjects to discriminate tones into high or low pitches with vocal responses
(audio-vocal responses); in task 2 subjects watched visually presented characters
and performed a choice reaction time task with manual responses (visual-motor re-
sponses). They found that practice dramatically reduced dual-task interference in
PRP.
The basic PRP effect has been modeled by several major computational cognitive
models based on production rules, notably EPIC [31] and ACT-R/PM [7]. Based on
its major assumption that production rules can fire in parallel, EPIC successfully
modeled the basic PRP effect by using complex lock and unlock strategies in cen-
tral processes to solve the time conflicts between perceptual, cognitive, and motor
processing [31]. However, neither EPIC nor ACT-R/PM modeled the practice effect
on PRP.
Here we modeled PRP effects with the same model that modeled typing phenom-
ena and integrated queuing network theory [26, 27] with reinforcement learning al-
gorithms [46]. Model simulation results were compared with experimental results of
both the basic PRP paradigm and the PRP practice effects [49]. All of the simulated
human performance data were derived from the natural interactions among servers
and entities in the queuing network without setting up lock and unlock strategies or
drawing complex scheduling charts.
9.3.1 Modeling the Basic PRP and the Practice Effect on PRP
with Queuing Networks
Figure 9.5 shows the queuing network model that was used to model PRP effects.
The model architecture is identical to the model that was used to model typing
