Connectionists: The Atoms of Neural Computation

[Brad Wyble]

>
> I love a discussion here on the group,  especially re: the Table of
> possible computation and their neural realizations, in the Supplement. We
> plan to crowd-source a more detailed version of that table; please contact
> me if you are interested in contributing.
>
>
For this I'd like to reinforce Stephen Grossberg's point about paying close
attention to behavioral data, which in my opinion should play at least as
strong a role as the neural data in our search for suitable neural
mechanisms to describe cognitive function.

For example if one is developing a model of working memory,  a set of
behavioral constraints provides strong restrictions on the classes of
mechanisms that one should consider which may be even more restrictive than
highly detailed neural data. As an example, we've recently published a
computational model (Swan & Wyble 2014, linked below) of binding
multi-feature objects into visual working memory that relies on satisfying
several general constraints, as well as matching certain empirical
benchmarks. These constraints are:

1. Memory can store repetitions of the same item
2. Memory is addressable by content and by temporal order
3. Storing multiple items produces interference and crosstalk
4. Items can be stored and (mostly) overwritten quite rapidly if the task
requires it

Obviously this list is far from complete, but even so it is useful in terms
of thinking about mechanisms. For example, the requirement to store
repetitions provides difficulty for models in which memory storage occurs
at the sensory level. I also suspect that repetitions pose difficulty for
some models that use neural synchrony to store binding information.

At the very least such behavioral constraints encourage one to think deeply
about how mechanisms would operate in service of functional requirements.

Best regards
-Brad

Swan, G., & Wyble, B. (2014). The binding pool: A model of shared neural
resources for distinct items in visual working memory. *Attention,
Perception, & Psychophysics*, 76(7) 2136-2157

*http://wyblelab.com/research_repos/models/bindingpool/
<http://wyblelab.com/research_repos/models/bindingpool/>*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20141104/c2e553c7/attachment.html>