R/P Distro

The "stimulus" is simply a trial defined to use the R/P Distro special operation. The "response sample" is one of four possible measures related to the subject's eye trajectory: the eye velocity vector magnitude; horizontal eye velocity; vertical eye velocity; and the direction of eye motion. In each of these cases, the response sample is averaged over the course of a specially designated segment in the trial, which is typically very short in duration. After examining the nominal response distribution collected in the assessment phase, the researcher defines a reward window spanning some portion of the nominal distribution and then resumes trial sequencing.

Again, the behavioral response is measured as before, but now the subject receives an enhanced reward ("pass") if the response falls within the reward window and a reduced reward ("fail") otherwise. The differential rewards are a function of the two reward pulses defined in the R/P Distro trial. Reward pulse #1 is the nominal reward delivered at the end of each trial during the assessment phase. Reward pulse #2 is only used during the modification phase and must be significantly shorter than pulse #1. During behavior modification, the "enhanced" reward is really two rewards: the shorter reward pulse #2 delivered immediately after the response is measured, plus the nominal reward pulse #1 delivered at trial's end. When the behavioral response falls outside the reward window, no reward is given after the response is measured, and the shorter reward is given instead at trial's end -- resulting in a "reduced" overall reward. Maestro again accumulates the response samples over many presentations of the trial, forming a modified response distribution. If the variability in the subject's behavior can indeed be altered by this approach, we would expect this second distribution to be skewed toward the defined reward window.

To conduct this kind of an experiment, you must first define a trial set that includes one or more trials employing the R/P Distro special operation and set the reward pulses appropriately for those trials. Maestro can track distribution and reward statistics for any number of such trials simultaneously. For example, one might define R/P Distro trials that measure eye velocity during the initial stage of pursuit of a constant velocity target moving in each of 8 different directions. Once you've entered Trial mode and selected the trial set in the Protocol tab, Maestro will catalogue all trials in the selected set that use the R/P Distro feature. Navigate to the R/P Distro tab page, and note that all such trials are listed in the dropdown combo box labelled Trial. Other controls on the dialog display additional relevant runtime information for the trial currently selected in the combo box; to look at a different R/P Distro trial, simply select its name from the dropdown list. Note that all controls are disabled while a trial sequence is running!

A small "canvas" window offers a graphical depiction of the "current" (white outline) and "previous" (filled in gray) response sample histograms collected for the trial; it will be blank if no response data has been collected. The horizontal extent of the canvas spans the valid response range defined in the numeric edit controls immediately above it. Specifying a reasonable response range may be useful, for example, if you want to ignore trials in which a saccade or an electronics artifact happens to occur during the special segment, resulting in an unreasonable value for the eye velocity. The histograms divide the valid response range into 25 bins. Samples falling outside the response range are still stored -- so if you later widen the range, the histogram will be adjusted to include the omitted samples. In the unnormalized display mode, the height of the histogram bars indicate the #counts in each bin; the vertical axis is scaled by the maximum observed bin count, which is written in the top left corner of the canvas (it is never less than 10). In the normalized display mode, the height of the histogram bars indicate the #counts in each bin relative to the maximum observed over that histogram; in this mode, the vertical axis spans [0.0 .. 1.0]. Simply click on the canvas to toggle between the two modes. Blue bars in the top and bottom margins of the canvas span the behavioral reward window for the trial, if it is enabled. If you want to enlarge this canvas, simply make the mode control panel larger; the canvas's right and bottom edges "stick" to the panel's corresponding edges.

As of Maestro v2.1.1, the experimenter can choose among four different behavioral response measures via the Response Type combo box. The available choices, as mentioned above, are all related to the velocity trajectory of the subject's eye. Note that changing the response type will automatically clear all previously collected response samples. Normally, the response type is selected prior to beginning the experiment.

The "response reward window" for the selected R/P Distro trial is defined by controls in the Reward Window group box. Click the check box to enable or disable the reward window, and use the accompanying numeric edit controls to set the response range covered by the window. Note that reward windows may not extend beyond the valid response range; Maestro will correct invalid entries accordingly.

As of Maestro v1.5.0, the Shift By and UpdIntv parameters were introduced to permit auto-shifting of the reward window based upon the subject's measured responses during the reward phase of the protocol. Here's how it works in the case of a positive shift. After constructing the subject's nominal response distribution in the assessment phase of the experiment, the investigator would set the reward window [Rmin .. Rmax] skewed toward the high side of the distribution, set Shift By to some small positive number δ, and set UpdIntv to the number N_upd of valid response samples collected per window update. After N_upd valid responses are collected, Maestro computes the mean response R. If R > Rmin, then the subject's response has "shifted" in the right direction, and Maestro shifts the reward window to [Rmin+δ .. Rmax+δ]. This process repeats until the investigator stops the trial sequence. Analogously, if the goal is to shift the response distribution to the left, the initial reward window would be skewed to the low side of the nominal distribution and δ would be set to a small negative number. In this case, Maestro would shift the reward window only if the mean response R satisfied R < Rmax.

The Mean/Std Dev control group includes read-only static controls that display the number of valid samples collected (i.e., within the valid response range), the mean, and the standard deviation for the current and previous response sample distributions. Instead of displaying statistics across the entire set of valid response samples, the investigator may choose to include only the most recent samples by setting N to some postive number; if N is zero, all valid samples are included in the calculations. Note that N applies also to the histograms displayed in the canvas window, and that you can specify different values for N for the current and previous response distributions.

The Reward Stats control group displays pass/fail statistics accumulated during the modification phase of the R/P Distro protocol, both for the currently selected trial and across all R/P Distro trials that have been run since Maestro started up.

Clicking the Start New Distrib button will discard the previous distribution, make the current distribution the previous one, then reset the current distribution data. Clicking it twice will effectively discard all response sample data collected thus far.

Given the controls and readouts on the R/P Distro tab page, running an experiment that aims to modify the variability in a subject's behavioral response is straightforward. Switch to Trial mode and select the trial set that contains the R/P Distro trial or trials you wish to present. Before starting the trial sequence, switch to the R/P Distro tab and, for each R/P Distro trial in the set: (1) select the type of behavioral response to be measured; (2) initialize the valid response range; (3) make sure the reward window is disabled; and (4) discard any previously collected response samples by clicking the Start New Distrib button twice. Then start trial sequencing to begin the assessment phase, watching the nominal response sample distribution build up for each relevant trial. When sufficient data has been collected, pause the trial sequence and use the controls in the Reward Window group to set up the response reward window for each trial. Also, click the Start New Distrib pushbutton once for each trial to make the "current" distribution the "previous" one. Then resume trial sequencing to start the modification phase, and keep the R/P Distro dialog in front to monitor the new response distributions and pass/fail statistics as they are collected by Maestro.

When you're done, you will probably want to save a summary of the statistics that have been collected during the experiment. Simply click on the Save Summary pushbutton and select a destination file. Maestro will write a summary of the runtime information collected thus far, in plain text format. As you can see in the example output file at the bottom of this page, the summary includes the overall pass/fail reward statistics and, for each R/P Distro trial:

• • The name of the trial.

  • The type of behavioral response measured (a string that matches the items in the Response Type combo box).

  • The valid response range assigned to the trial.

  • The reward window parameters, if the window is enabled.

  • Pass/fail statistics for the trial.

  • All samples collected for both the current and previous distributions, as well as sample histograms calculated over the valid response range.

Note that there is a utility available, getrpdsummary(), that will read the contents of this file into Matlab.