| |
Abstract:
Abstract: We propose a cognitive architecture based on three
weak assumptions. First, cognition can be characterized by the
operation of a relatively small set of information processing
pathways. Second, the pathways are dynamically interconnected based
on task demands. Third, each pathway exhibits a speed-accuracy
trade off. Based on these assumptions, we explore the temporal
characteristics of information transmission through a series of
pathways. We model a pathway as a temporal belief network, which
encodes and updates probability distributions over states and time.
We examine the role of input ambiguity, state similarity, number of
alternative states, and the arrangement of pathways on the behavior
(overall speed and accuracy) of the model. Our goal is to
understand universal computational properties and limitations of
any cognitive system that satisfies our weak assumptions. This work
expands on the seminal "cascade model" of McClelland (1979), but
the mathematically principled framework of a temporal belief
network allows for stronger conclusions and the ability to explore
a broader range of issues.
|
