[ACT-R-users] Set-similarities.

[Laura Hiatt]

We recently modeled similarity between perceptual objects using perceptual similarity (such as the similarity of color RBG values) and activation:

http://mindmodeling.org/cogsci2013/papers/0126/paper0126.pdf

It matches experimental data quite well and was learned over time (at least the BL / spreading activation part was - the perceptual similarity is currently static).  We're still actively working on this - happy to chat about it.


Another paper that might be of use is:

http://www.aaai.org/ocs/index.php/WS/AAAIW11/paper/viewFile/3935/4300

which uses ACT-R in conjunction with Leabra to classify visual objects. 

Best,
Laura


On May 9, 2014, at 9:04 PM, db30 at andrew.cmu.edu wrote:

> 
> 
> --On Friday, May 09, 2014 05:03:36 PM -0700 Ganapathy Priya <gapri13 at gmail.com> wrote:
> 
>> Dear ACT-R users,
>> 
>> We are working on modeling a task of visual object recognition. We
>> would like to update the similarities of chunks (objects) as the model
>> learns to recognize objects. We would like to start by making the
>> set-similarities of chunks as 0 and then update the chunk
>> (dis)similarities in small increments from 0 to -1 across several
>> training rounds.
>> 
>> Our chunks look like this: (add-dm (Caa1 isa object-content image "aa1"))
>> (add-dm (Cab1 isa object-content image "ab1"))
>> 
>> (set-similarities  (Caa1 Cab1 0))
>> 
>> We are looking for sample codes that anybody can share where they have
>> updated the chunk similarities during run-time. Any other functions to
>> alter the similarities will also be useful.
>> 
> 
> Set-similarities/set-similarities-fct is the only recommended command for
> setting the similarites between chunks, but sdp/sdp-fct can also be used to
> do so (see the reference manual for details).  Either of those commands
> can be called at anytime to change similarites.  Alternatively, one can set
> the :sim-hook parameter to specify a function which is used to compute
> similarities "on the fly".  The assignment model in unit 5 of the ACT-R
> tutorial uses that to set similarities between non-chunk values, but it can
> also be used to dynamically compute chunk similaritles.
> 
> Hope that helps,
> Dan
> _______________________________________________
> ACT-R-users mailing list
> ACT-R-users at act-r.psy.cmu.edu
> https://mailman.srv.cs.cmu.edu/mailman/listinfo/act-r-users