EASE: Learning and Sensory-based Engagement, Arousal and Self-Efficacy (EASE) modeling
This NSF SCH funded project is a joint effort between UCCS Computer Science (T. Boult PI and R. Lewis), UCCS Psychology (Charles Benight), CMU CS (F. De La Torre) and U. Pitt Psychology (J. Cohn).
Mental Trauma following disasters, military service, accidents, domestic violence and other traumatic events, is a health issue costing multiple billion of dollars per year. Beyond its direct costs, there are indirect costs including a 45-150% greater use of physical medical care. While web-based support systems have been developed these are effectively a “one-size-fist all” approach lacking the personalization of regular treatment, and the engagement and effectiveness associated with the tailored regiment.
A multi-disciplinary team of leading researchers in trauma treatment, facial analysis, computer vision and machine learning, will develop a scalable adaptive person-centered approach that uses vision and sensing to measure Engagement, Arousal and Self-Efficacy (EASE) during treatment and to adapt treatment. We build on well established social-cognitive theory, where arousal and self-efficacy are critical elements of recovery. We measure engagement, critical in self-directed web-based treatment. We will show that sensing can meaningful improve treatment effectiveness. Building a smart system that empowers individuals by combining sensing and learning to improve treatment offers a transformative approach this national health need.
This effort is the first research on social cognitive theory seeking to operationalize self-regulation through face, voice, and physiological arousal within sessions of a web-intervention system. It will be the first to attempt to infer changes in self-efficacy, during treatment, from sensing/observation.
Underlying the health application is the need for fundamental advances in vision, sensing and facial analysis as well as advances in machine learning. This work will develop and evaluate new approach for directly estimating engagement and arousal. Being able to estimate physiological response, heart rates and signs of arousal, directly from ubiquitous face and voice data requires advances in personally adapted modeling and experimental assessment to evaluate the effectiveness.
This effort will address important problems in domain adaption for missing/noisy variables. Individuals have different states of recovery with different issues and responses to stimuli, so training will develop multiple models. Because the ground truth will often only have partial estimates of EASE variables, requiring advances in manifold learning with noisy/missing data. For effective trauma treatment, the appropriate model will need to be selected and rapidly adapted to the user during the early measurements, again requiring advances in the state of art in domain adaption.