MENTAL WORKLOAD ASSESSMENT WITH NEURAL NETWORKS

NEURAL NETWORKS WILL BE USED TO DEVELOP A REAL-TIME, MULTIMODALITY, NONINTRUSIVE PHYSIOLOGICAL METRIC OF MENTAL WORKLOAD. OUR METRIC WILL MEASURE MENTAL WORKLOAD RELATIVE TO TWO OR MORE CALIBRATION POINTS FOR A FAMILY OF CLOSELY RELATED TASKS WHERE MENTAL LOADING IS THE MAIN VARIABLE DIFFERING BETWEEN TASKS. THE METHODS WILL BE BASED ON PHYSIOLOGICAL SIGNALS THAT NEED NOT BE EVOKED BY IRRELEVANT STIMULI AND DO NOT REQUIRE THE USE OF ARTIFICIAL TASK STRUCTURES. DURING PHASE I, WE WILL TEST FEASIBILITY BY IDENTIFYING COMBINATIONS OF EYE BLINK, HEART RATE VARIABILITY AND REGIONAL NEUROELECTRIC SIGNAL FETURES THAT WILL BEST DISTINGUISH BETWEEN TWO MENTAL WORKLOAD LEVELS OF A VISUOMOTOR JUDGEMENT TASK PREVIOUSLY RECORDED FROM FIVE USAF FIGHTER TEST PILOTS. WE WILL USE FEED-FORWARD NEURAL NETWORKS, RECURRENT UNSUPERVISED NEURAL NETWORKS, AND BAYESIAN CLUSTER ANALYSIS TO ACHIEVE THIS GOAL. DURING PHASE II, WE WILL TEST THE SPECIFICITY AND GENERALIZABILITY OF THE METHOD BY TESTING IT ON DATA RECORDED IN FLIGHT SIMULATORS AND DURING PERFORMANCE OF SEVERAL STANDARDIZED COGNITIVE TASKS. WE WILL ALSO INCORPORATE OTHER PASSIVE PSYCHOPHYSIOLOGIAL MEASUREMENTS SUCH AS ELECTRODERMAL ACTIVITY AND TONIC EMG, REFINE THE METHOD, AND DESIGN A DEVICE THAT CAN BE USED TO CONSTRUCT THE WORKLOAD INDEX FOR ANY OPERATOR AND ANY FAMILY OF TASKS.