Specifications for Moving Target Fractions Software

We created a preliminary design for an adaptive information-filtering algorithm. In target-selection tasks, when the rate of new incoming targets becomes too fast for the user to handle, the user may possibly benefit from computer-aided pre-selection of preferred targets. If the selection of targets requires human judgment (otherwise a completely automated system without any humans would work best), but a computer can approximate this judgment although not to the accuracy of a human, then the computer could possibly help the user by pointing out the targets that it believes to be the greatest threats. The cognitive load of judgment can be simulated in an experiment by requiring the user to perform some arbitrary cognitive task to determine whether a target should be selected such as comparing fractions (e.g., mousing over a target or clicking on a target could reveal a fraction, which if over 1/3 means this target should be selected). If there are negative targets (equivalent to friendly units in a battlefield), then the user cannot just select all visible targets to maximize performance. Also, by adding variable point values to positive (enemy) targets, the performance of the human-machine system can be measured objectively by adding the total points of selected enemy targets minus the total points of selected friendly units during an experiment session. If the rate of new incoming targets is too fast for the user to evaluate all of the new targets, then the computer can help to improve the system performance by annotating the best targets. If the computer varies the threshold of annotation based on the user's currently measured cognitive load, then it may be possible to better the performance of a fixed threshold system. Depending on the exact point assignments, the target order-of-magnitude improvement may be possible. This experiment has the advantage that it is clearly relevant to the DoD since target selection is an important part of many military tasks.

Here are more specific requirements:

For prototype 1, we need:

• Each session will have a variable numbers of targets, increasing and decreasing over time according to preset parameters. Initially, it may be fine to have a fixed number of targets per session (each session lasting a few minutes - the exact number of minutes to be specified later) evenly distributed across the session. The session officially ends when the last target leaves the screen.
• Targets move across the screen during each session. Initially, targets just move from left to right from a random point along the left edge straight to a random point on the right edge. Future improvements might include random Brownian paths between those two points. Depending on experimental design requirements, we may want to precompute the paths and just store them in a table for each session.
• Targets must be selectable by left clicking the mouse and deselected by right clicking the mouse. Selecting or deselecting a target makes the target disappear. Future improvements might include graying out the target.
• Targets display a cognitive task that the user must perform to decide whether to select the target. Initially the task will be to determine whether a fraction displayed is above a certain value, 1/3. Displays will be predetermined and stored in a table for each session.