Instructions for using the demo version of the IAC model

Click here to start up the model on your computer

The program here is a demo version of the interactive activation and competition (IAC) model of face recognition.  This is a simulation developed as part of our group's explorations of face recognition over a number of years.  You can find the most up to date description of the model in

Burton, A.M., Bruce, V. & Hancock, P.J.B. (1999). From pixels to people: a model of familiar face
         recognition. Cognitive Science, 23, 1-31.  ( PDF here ,  1.8MB)

The original report, setting out the basic effects in the model, is

Burton, A.M., Bruce, V. & Johnston, R.A. (1990). Understanding face recognition with an interactive
         activation model. British Journal of Psychology, 81, 361-380.

You can find lots more references to work with the model  here

This simple version of the IAC model is written in Java, and should run on any computer, as long as the browser you are using is Java enabled.

These instructions are in two parts, first how to run the program, and what the various parts do.  Second, there are some pointers to potentially interesting investigations you can carry out with the model.

NOTE: This is a necessarily stripped-down version of the model, and is intended to demonstrate the main concepts behind it.  It is not an exact replica of the model used for research purposes, which is slightly more complicated.  If you need details of the original model, see the various appendices in papers describing it, or contact Mike Burton.
 

Basic usage

When you first see the model, the "show activations" button on the top right is pressed.  This means that what you are looking at is the activation levels of all the units.

Activation:  black = negative;  white = positive.  Size of unit is the amount of activation.
    You will notice that at the start, all units have a small negative activation, this is the resting level.

Click on a unit to cause external activation to be passed to that unit.  You will see a square to remind you which units are receiving external activation.   You can click again on the same unit to turn external activation off.  Normally, you would only provide external activation to the FRUs or NRUs.

The Step button causes the model to cycle (i.e. update all it's unit activation levels) once.  Press it many times to watch activation flowing through the system.  (You will only notice anything happening if you have provided external activation to at least  one unit.)

The Number button controls how many times the models cycles each time you press Step.

The Cycles field tells you how many cycles you have run.

The Reset button resets all the unit activation, and turns off any external activation.

The Show links button changes the mode of the simulation.  Now you see the links to any unit, not the activations of units.  Click on a unit to see all the units in the model which are linked to it.  Once again, black is a negative link, white a positive.  What you should see is that the model has within-pool inhibition,  and that particular units are connected across pools with excitatory (positive) links.

This has explained the basic mechanics of the model.  The next section describes some possible projects for understanding how it has been used in face recognition research.
 

Some projects

1. Dynamic properties
Click on an FRU and cycle the model 50 or so times.   You should see activation passing from that FRU on to the appropriate PIN, and then on to the SIUs (which code information about people).  After a while, (i.e. number of cycles) the model will settle, and no further activation change will take place.  This is the point of equilibrium - the point at which the global decay and all the inhibitory forces on units are matched by the excitatory forces on units.
Now click on the FRU again, to turn external activation off.  If you cycle the model many times, you should see that the activation will slowly die away in all units, though this takes a long time to achieve.
Now repeat the process.  Activate an FRU and cycle the model until it reaches a stable state.  Thit time, when you turn off the FRU, turn another one on at the same time, and cycle the model.  What do you see?  You should see that the new FRU causes its relevant PIN to become active,  but furthermore,  it should have the effect of forcing down activation in the previous person much faster than normal decay.   In this way, competition between units makes the model dynamic.  You don't have to "wipe it clean" with a reset every time you introduce more stimuli.  Instead,  it is a natural property of the model that each new representation that is activated tends to erase activation of recently-presented stimuli.

2. Semantic priming
Activate an FRU and run the model until it settles.  Sometimes when you do this, you will see that PINs other than the person activated become active.  Can you work out when this happens and when it doesn't?   In fact,  when two PINs share many SIUs,  activation of one causes activation in the other.   Now turn off the active FRU, and instead activate the FRU corresponding to a slightly-active PIN.  Notice that this PIN rises much faster than normal.
This is the model of semantic priming we have offered in the papers above (and others).  As an exercise, try to answer these questions about semantic priming by experimenting with the model.
A.  Semantic priming experiments show that it "crosses domains", i.e. names prime faces just as much as faces prime faces. How is this explained within the model?
B. Semantic priming disappears after a relatively short period - particularly if the period is filled with an unrelated face.  Why?

3. Interference
One of the interesting features of this model is that many units can take activation at the same time.  Try activating a person's FRU at the same time as activating their NRU.  What are the effects? How does this compare with the single-route case?
Many of the studies in face recognition use an interference procedure, in which names and faces are mis-matched, while subjects are asked to make a semantic decision (e.g., politician/pop-star?)  which is shared by one or both of the people.  Can you derive some predictions about the pattern of interference you would expect from experiments such as these?
 


Back to Research in Face Recognition
Back to Department of Psychology Homepage