Machine Learning in Medicine Virtual Seminar
Speaker
IOANNIS T. PAVLIDIS
Eckhard-Pfeiffer Professor of Computer Science, and
Director, Computational Physiology Laboratory
University of Houston
When
-
Where
Virtual Presentation - ET
Description
In present human-machine partnerships, the relationship between drivers and cars looms large; it defines the daily lives of billions of people, it has broad safety implications, and carries longterm health consequences. The drivers’ parameters in this relationship also stand to inform the transition to increased levels of vehicle automation, where machines need to acquire a deeper understanding of the human co-drivers. For the last 10 years, our lab has systematically studied driving behaviors in vehicles with Level 2 (L2) automation. Our investigations progressed from controlled lab studies, to controlled race-track studies, to semi-controlled on road studies, and eventually to naturalistic driving studies. Across this spectrum of realism, our research focused on the genesis and role of micro-stressors in human-machine interactions. We documented the following consequential behavioral patterns: a) Sixth Sense Driving, where a conflict resolution mechanism in the anterior cingulate cortex (ACC) subconsciously manages driving performance during mental distractions. The ACC mechanism degrades ungracefully in the presence of physical distractions, such as texting while driving, which explains the potent risk such distractions pose. b) Accelarousal, where mundane acceleration, such as in stop and go traffic, stimulates hyperarousal events in a significant minority of the driving population. The resulting hyperaroused states have mean intensity ~50% over the baseline, featuring all the characteristics of a long-term hidden stressor.
In response to these findings, we designed and tested machine learning (ML) methods that predict distracted and acceleration-laden driving in near-real time, alerting drivers about negative habitual behaviors that have long drifted into their subconscious. We also designed and tested biofeedback methods that arrest hyperarousal events during driving, thus ameliorating their long-term impact. These are examples of affective correction, providing machines with the means to help humans rationalize their driving styles, when the latter are in control of L3 and L4 vehicles. Such interactions may seem exotic right now, but they will be necessary in a mixed human-machine driving world, where driving standards between the `two species’ would need to be homogenized.
To carry out all these studies, we developed state-of-the-art affective computing methods, with the capacity to acquire and wrangle vehicular, physiological, observational, and psychometric data in near real-time. Recently, we have launched naturalistic ubiquitous (NUBI) driving studies, where drivers around the state of Texas are being monitored unobtrusively as they go about their daily activities. In the NUBI studies, contextual information about traffic and weather conditions is extracted from Internet sources and matched to the other data channels, providing a 360o view, ripe for causal reasoning.
About the Speaker.
Zoom Participation. See announcement.