From Computational Cognitive Neuroscience Wiki

Self Regulation: Accommodation and Hysteresis

• The project file: self_reg.proj (click and Save As to download, then open in Emergent)

Back to CECN1 Projects

Project Documentation

(note: this is a literal copy from the simulation documentation -- it contains links that will not work within the wiki)

• To start, it is usually a good idea to do Object/Edit Dialog in the menu just above this text, which will open this documentation in a separate window that you can more easily come back to. Alternatively, you can always return by clicking on the ProjectDocs tab at the top of this middle panel.

This project looks pretty much like the unit.pro simulation you explored earlier, except that it plots more variables (the basis and conductance values for accommodation and hysteresis), and contains more parameters to control these self-regulatory channels in the control panel. On the .PanelTab.ControlPanel, you should notice at the bottom that there are on_cycle and off_cycle parameters for two "stimuli", and the overall run is for 500 cycles.

• Click on the .T3Tab.CycleOutputData tab to view the graph, which now contains new variables, in addition to net and act_eq. Go ahead and Init and Run.

You should see that the activation result (without either accommodation or hysteresis turned on yet) is just as one would expect -- a bump at 10 cycles lasting until 100, and another from 300 to 400. Note in particular that there is no apparent effect of the prior activation on the response to the latter one.

Now, let's turn on accommodation.

• Click the on! button in the acc field (which contains the accommodation channel parameters) of the control panel and press Apply. Then, in the graph view window. Press Run again.

You should observe that the vcb.acc line in the graph increases when the unit becomes active, but that it does not quite reach the activation threshold, which is set to .7 (see a_thr in the acc field of the control panel).

The unit needs to remain active for a little bit longer for the basis variable to accumulate to the point of activation. One way we could achieve this is to set the offf_cycle1 time for the first stimulus to be somewhat later.

• Set the off_cycle1 to 150 instead of 100, and then press Run again.

Now you should observe that the gc.a accommodation conductance starts to increase just before the stimulus goes off. Even though the unit becomes inactive, it takes a while for the basis variable to decrease to the deactivation threshold (.1, in d_thr of the acc field). Indeed, it takes so long that by the time the next input stimulus comes in at 300 cycles, there is still a strong accommodation current (gc.a), and the unit is not immediately activated by the input. Only at around 350 cycles, when the basis variable finally goes below the deactivation threshold, and thus gc.a starts to decrease, does the unit then become active.

Had this unit been in a network with other units that received input from the second stimulus, but some of these were not activated by the first stimulus, then these other units would get active immediately because they were not "fatigued," and they would produce a different representation of the input. Thus, accommodation provides one means for a network to respond differently to subsequent inputs based on prior activity.

At this point, you can explore the various parameters in the acc field and other associated parameters to make sure you understand what they do.

• For example, play with g_bar.a, which determines the strength of the overall accommodation current. Change g_bar.a from .5 to .3, and you should see that this is too weak to completely prevent the unit from immediately becoming active when the second stimulus comes in at 300 cycles.

Now, let's add hysteresis into the mix.

• First press the Defaults button to return to the default parameters. Then click on the hyst.on! button on, and do Run.

You should see that the unit remains active for some time after the input stimulus goes off at 100 cycles --- this is caused by the hysteresis. Further, this additional period of activation causes the accommodation current to get activated, which eventually turns the unit off. As a result of this accommodation, as we saw before, the unit does not then get activated immediately by the second input.

• Now, you can play with the hyst parameters and see what they do to the unit's response properties.

In summary, you can see that there is considerable potential for complex dynamics to emerge from the interactions of these different channels. The fact that actual neurons have many such channels with even more complex dynamics suggests that our basic point neuron model is probably a lot simpler than the real thing. Due to pragmatic constraints, we typically ignore much of this complexity. As the simulations later in the text demonstrate, such simplifications do not make that much of a difference in the core aspects of behavior that we are modeling.