|
|
|
The aim of the symposium is to compare and contrast two views on mind
and language that guide much of contemporary research in cognitive
science. One view, connectionism, sees the important aspects of
cognition as emerging directly from the joint activity of a large
network of individual processing units. The second view, the
structured-Bayesian approach, sees cognition as involving a special
kind of inference in complex probabilistic models.
The views often seem to make genuinely conflicting claims about how we should think about cognition and language. Are the two approaches simply aimed at different levels of analysis? If so, should we see them as explanatory in different ways? And how could they eventually fit together? Can Bayesian algorithms be implemented in connectionist models? Or should connectionist models be seen as a particular kind of probabilistic model that perhaps only approximates Bayesian inference? In that case, what do we lose by abstracting away from the neural mechanisms and considering the idealized models in isolation? Or conversely, what do we gain by designing models that adhere to strict Bayesian principles? Is approximation to these models the best way of understanding the computations neural networks are performing? Our goal is to address these and other questions by bringing together several top researchers from both approaches to offer different views on how we might begin to answer them. |