Michelle G. Newman - US grants

The emergence of massive human-rated and commented visual data has opened avenues for exploring fundamental questions in artificial intelligence beyond the horizon. This project tackles the challenge of automatically inferring visual aesthetics and emotions and inventing new systems that assist creative and decision-making activities of the general public. An interdisciplinary team, with expertise in visual modeling, data mining, psychology, and computational sciences will build tools to distill information from a combination of visual, textual, and numerical data. Visual features, selected based on published literature and consultation with domain experts, will be extracted for discriminating types of emotions. The resulting systems can select and rank visual information based on aesthetics and emotions.

Intellectual Merits: This project will allow computer scientists to gain understanding of next-generation computerized visual aesthetics and emotion assessment systems. The complex inter-relationship among content, context, and subjectivity in aesthetics and emotion assessment makes the corresponding learning problems especially challenging, which is likely to trigger innovation in machine learning and statistical modeling. Such capabilities will fundamentally change the way visual information is analyzed, processed, and managed. The project will advance our understanding of the computability of emotions, and lead to new applications that can be used in a variety of settings.

Broader Impacts: The research will have a transformative impact in the fields of information retrieval, human-computer interaction, information processing, consumer electronics, and design. The technology can also be used to refine multimedia content that serves as education resources. The project will disseminate research findings, generate new software implementations and collected datasets, and provide online services that can be used by researchers, educators, and industry. Education efforts include developing an interdisciplinary curriculum, training cross-disciplinary scientists, and involving underrepresented groups in research.

PROJECT SUMMARY/ABSTRACT The prevalence of mental health problems among college populations has risen steadily in recent decades, with one-third of college students struggling with anxiety, depression, or an eating disorder. Yet, only 20-40% of college students with mental disorders receive treatment. Inadequacies in mental health care delivery result in prolonged illness, disease progression, poorer prognosis, and greater likelihood of relapse, highlighting the need for a new approach for detecting mental health problems and engaging college students in services. We have developed a transdiagnostic, low-cost mobile health targeted prevention and intervention platform that uses population-level screening for engaging college students in tailored services that address common mental health problems. This care delivery system represents an ideal model for service delivery given its use of our promising, evidence-based mobile programs, a transdiagnostic approach that addresses comorbid mental health issues, and personalized screening and intervention to increase service uptake, enhance engagement, and improve outcomes. Further, our service delivery model harnesses the expertise of our team of leaders in behavioral science, college student mental health, technology, and health economics, and bridges our team's work over the past 25 years in successfully implementing a population-based screening program in over 160 colleges and demonstrating the effectiveness of Internet-based programs for targeted prevention and intervention for anxiety, depression, and eating disorders in over 40 colleges. We propose to test the impact of this mobile mental health platform for service delivery in a large-scale trial across 20 colleges. Students who screen positive or at high-risk for clinical anxiety, depression, or eating disorders (excluding anorexia nervosa, for which more intensive medical monitoring is warranted), which account for a substantial proportion of the mental health burden on college campuses, and who are not currently engaged in mental health services (N=7,884; of 146,000 initially screened) will be randomly assigned to: 1) intervention via the mobile mental health platform; or 2) referral to usual care (i.e., campus health or counseling center). We will test whether the mobile mental health platform, compared to usual care, is associated with improved uptake (i.e., individuals beginning treatment) (Aim 1), reduced clinical cases and disorder-specific symptoms (Aims 2a, 2b), and improved quality of life and functioning (Aim 2c). We will also test putative targets/mechanisms, other mediators, predictors, and moderators of improved mental health outcomes (Aim 3) as well as stakeholder- relevant outcomes, including cost-effectiveness and academic performance (Aim 4). Our comprehensive mental health care platform can yield clinical benefit to students, appeal to university stakeholders, minimize barriers to implementation sustainability on campuses, and produce an economic return on investment compared to usual care. This population-level approach to service engagement has the potential to improve mental health outcomes for the 20+ million students enrolled in U.S. colleges and universities.

This CISE Community Research Infrastructure (CCRI) planning project aims to form a community of researchers galvanized around the creation of a very large scale, high-quality annotated video dataset for broad use by the affective computing community, especially for computational understanding of human bodily expressions in the wild. Computational understanding of body language for emotional expressions in real-world environments is a fundamental and challenging research topic. Breakthroughs in this technological area have the potential to enable a large number of innovative applications including personal assistant robot, social robot, and multimedia information retrieval. The planning project will encourage the scientific community to take a multidisciplinary and holistic approach to the problem of body language understanding. The project aims to gain deeper insights into useful data collection for computational and data-driven modeling of emotions based on body language. Ultimately, by analyzing and harnessing a large quantity of carefully-collected affective data, using state-of-the-art machine learning and statistical modeling methodologies, future intelligent systems can be empowered with new capabilities of understanding human emotions.

This planning project will involve researchers in computer and information sciences, social and clinical psychology, statistics, and data ethics. A workshop will be organized in conjunction with one of the leading international conferences in the research area. The team will reach out to relevant communities to attract researchers to participate in the workshop. At the workshop, a data-driven emotion modeling competition will be organized using a preliminary dataset that the team has collected. Winning teams will be invited to present their results at the workshop. The workshop will also feature panels and discussion sessions to solicit input from the broader community on the design and implementation of the potential data infrastructure. The planning effort will result in a publication on the consensus reached at the workshop on the community's need for this data infrastructure and the detailed specifications on various aspects of the data collection effort.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.