activation and latency recommendations for 5.0

Tue Jan 15 16:48:42 EST 2002

Content-Type: text/plain; charset="us-ascii" ; format="flowed"

The main point of this message is to recommend that users of 5.0 
adopt activation settings and chose the latency equation that will 
make ACT-R 5.0 more closely correspond to the behavior of 4.0 with 
respect to these subsymbolic computations.  The following is our 
recommendations for the setting of the goal-activation and latency 
equation:

(sgp :esc t :ga 1.0)

(setf *latency-fn* 'old-latency)

Except for the fact that retrieval threshold is variable, this will 
make the behavior of 5.0 identical to 4.0.  It is the setting that we 
will use for the 5.0 tutorial and which will serve as the basis of 
this year's summer school.  Note, in particular, setting ga to 1 will 
leave the Sji's behaving as they use to and setting the latency 
function will replace competitive latency with the old latency 
equation.  It is not recommended, however, that you turn associative 
learning on and we do not regard the learning equations as part of 
the 5.0 theory.

Of course, in the ACT-R community we always allow a 1000 flowers to 
bloom and you should feel free to explore other options (including 
competitive latency and associative learning or variants on these) 
and report to us what you have learned.

In brief our assessment of this is that, while competitive latency 
had real motivations, its current instantiation introduced a set of 
problems.  We were not able to get all of the old 4.0 models to 
convert under this setting and this violated the constraint of 
cumulative progress.  The few who were working with competitive 
latency were encountering new problems.  A somewhat more elaborate 
discussion of the issues behind this recommendation is available at 
http://act.psy.cmu.edu/ACT-R_5.0/activation_and_latency.doc.
-- 

==========================================================

John R. Anderson
Carnegie Mellon University
Pittsburgh, PA 15213

Phone: 412-268-2788
Fax:     412-268-2844
email: ja at cmu.edu
URL:  http://act.psy.cmu.edu/
--============_-1200991171==_ma============
Content-Type: text/html; charset="us-ascii"

<!doctype html public "-//W3C//DTD W3 HTML//EN">
<html><head><style type="text/css"><!--
blockquote, dl, ul, ol, li { margin-top: 0 ; margin-bottom: 0 }
 --></style><title>activation and latency recommendations for
5.0</title></head><body>
<div><font face="New York" color="#000000">The main point of this
message is to recommend that users of 5.0 adopt activation settings
and chose the latency equation that will make ACT-R 5.0 more closely
correspond to the behavior of 4.0 with respect to these subsymbolic
computations.  The following is our recommendations for the
setting of the goal-activation and latency equation:</font><br>
<font face="New York" color="#000000"></font></div>
<div><tt><font color="#000000">(sgp :esc t :ga 1.0)</font></tt></div>
<div><tt><font color="#000000"><br>
(setf *latency-fn* 'old-latency)<br>
<br>
</font></tt><font face="New York" color="#000000">Except for the fact
that retrieval threshold is variable, this will make the behavior of
5.0 identical to 4.0.  It is the setting that we will use for the
5.0 tutorial and which will serve as the basis of this year's summer
school.  Note, in particular, setting ga to 1 will leave the
Sji's behaving as they use to and setting the latency function will
replace competitive latency with the old latency equation.  It is
not recommended, however, that you turn associative learning on and we
do not regard the learning equations as part of the 5.0
theory.</font><br>
<font face="New York" color="#000000"></font></div>
<div><font face="New York" color="#000000">Of course, in the ACT-R
community we always allow a 1000 flowers to bloom and you should feel
free to explore other options (including competitive latency and
associative learning or variants on these) and report to us what you
have learned.</font></div>
<div><font face="New York" color="#000000"><br></font></div>
<div><font face="New York" color="#000000">In brief our assessment of
this is that, while competitive latency had real motivations, its
current instantiation introduced a set of problems.  We were not
able to get all of the old 4.0 models to convert under this setting
and this violated the constraint of cumulative progress.  The few
who were working with competitive latency were encountering new
problems.  A somewhat more elaborate discussion of the issues
behind this recommendation is available at</font>
http://act.psy.cmu.edu/ACT-R_5.0/activation_and_latency.doc<font
face="New York" color="#000000">.</font></div>

<div><font color="#000000">-- <br>
</font></div>
<div><font
color="#000000"
>==========================================================</font></div
>
<div><font color="#000000"><br></font></div>
<div><font color="#000000">John R. Anderson<br>
Carnegie Mellon University<br>
Pittsburgh, PA 15213<br>
<br>
Phone: 412-268-2788<br>
Fax:     412-268-2844<br>
email: ja at cmu.edu<br>
URL:  http://act.psy.cmu.edu/</font></div>
</body>
</html>
--============_-1200991171==_ma============--