Documentation for the Episodic Simultaneous Type, Serial Token Model (or eSTST) of visual working memory.   

This computational model is described in Wyble, Bowman & Nieuwenstein (In press at JEP:HPP).  The model is a neural network that takes as input a temporal arrangement of targets in an RSVP stream and simulates the process of encoding those targets into working memory.  The output of the model includes the identity, and temporal ordering of those targets, including repetitions.  

The matlab code included here implements the model and includes a graphical user interface, as well as command-line scripts.  


Graphical Interface:

Executing the script Runmodel.m opens panels that allows simulation of single RSVP trials or experiments.


Command Line Interface:

Executing batchrun(0) simulates a number of conditions by calling eSTSTbattery.m.  Use an argument of 1 to create a progress bar.

Executing eSTSTbattery(n, 0) steps through a series of trials to simulate one block of an experiment in a prototypical subject.
n specifies the set of conditions to simulate (see setlags.m)
This function also requires the global variables: SOA, bdelay, targetvariance, T1blank, base, range name and resolution  to be specified.  See batchrun.m for example settings.

Executing eSTSTmodel(numerous parameters)  simulates a single trial in each condition.  See the .m for clarification of the parameters required.  

Executing trialeval(n) evaluates the performance of a single set of trials of the model, returning what targets were encoded into working memory, in what order, and whether any repetitions were encoded.  Output is most easily observed in the variable condition.  

Executing analyzeblock evaluates an entire experimental block, based on data stored in the execution of eSTSTbattery.m

Output accuracy from analyzeblock is stored in the global variable condition(c t), for each condition c, over the targets t.

Output order is stored in the global variable order(tnum,tpos,c) which stores the order of the targets in a 2D array for each condition c.  Each 2D arracy for a given condition specifies how often each target tnum was reported in position tpos.

Order computation is currently configured for 2 targets.  For computing order among three or more targets, you will have to edit the analyzeblock.m file, and this analysis ignores trials on which targets are missed.


Notes:

The interface has been designed in Matlab release 7 (2007b), but the figures are intended to be compatible with releases 5 and 6.   I have not tested them on these earlier versions and would appreciate feedback at bwyble@gmail.com.  I have also not tested this code with Linux versions of matlab.  

The implementation listed here has been modified to allow up to 6 targets to be presented within a single trial.   There are now 6 tokens available in working memory.  

The activation traces shown for a single trial include the blaster, the input nodes and the type nodes.  Nodes in the binding pool are excluded from the predefined plots for clarity, but the arrays containing this information are stored, and well commented within eSTSTmodel.m

This is version current as on June 13, 2008.  

Check: 
http://www.bradwyble.com/research/models/eSTST/  

or email:
bwyble@gmail.com 

for updates.






