competitive learning (equation)?

[Dan Bothell]

--On Tuesday, July 23, 2002 4:25 PM +0100 Richard M Young 
<r.m.young at herts.ac.uk> wrote:

> Act-R folks,
>
> I feel a bit embarassed about asking what is presumably a rather basic
> question, especially with the Summer School going on and the workshop about
> to happen, but ... there's frequent mention in emails and workshop talks
> about something called "competitive learning" and its associated equation.
> The 1998 book doesn't have an index entry for it, and Table 12.2 of the
> basic equations doesn't have one thus called.
>
> Could someone explain briefly, or give me a pointer, to what "competitive
> learning" is?
>

Competitive latency was one of the big changes in the early versions of ACT-R 
5.0, and it has since been "discarded" as a bad idea for general practices, 
but there are some who still find it quite useful (it still exists in the 
ACT-R 5 code, but isn't enabled by default).  It replaced the retrieval time 
equation with one that made the time to retrieve a chunk dependent on not only 
that chunk's activation, but the activation of its competitors as well. 
Christian's subsymbolic component talk from last years post-graduate summer 
school discusses that system, and you can get the presentation from our site 
at: http://act.psy.cmu.edu/ftp/workshop/Workshop-2001/schedule.html.

Hope that helps,
Dan