Active Grants/Awards

Suicide is a public health concern and is ranked as the second leading cause of death in 10-24 years old.1,2 In particular, the increasing rates of suicide mortality and suicidal ideations and behaviors among Black youth in the United States (US) have become a pressing concern in recent years.3 Between 2001 and 2015, Black children under 13 years old were twice as likely to die by suicide, compared to their White counterparts.4 Furthermore, suicide mortality rates among Black youth have risen more rapidly than in any other racial or ethnic group.2,5 However, there remains a significant knowledge gap in understanding culturally tailored suicide prevention strategies for this population, particularly regarding unique social risk factors specific to Black youth. Specifically, a detailed understanding of social risk factors unique to Black youth and their differentiation from risk factors for other racial and ethnic groups is limited.2 This knowledge gap is critical, as research has indicated that Black youth face greater exposure to adverse childhood experiences (ACEs), which are linked to higher risks of suicidal ideation and attempts. The National Violent Death Reporting System (NVDRS) is a state-based violent death reporting system in the U.S. that helps provide information and context on when, where, and how violent deaths occur and who is affected.11 However, much of the information in NVDRS is unstructured, limiting its use in examining a complete picture of the social risk factors contributing to Black youth suicide. Therefore, it is imperative to develop machine learning (e.g., natural language processing [NLP]) algorithms to automatically extract social risk factors from free text to help analyze Black youth suicide. Our long-term goal is to reduce the suicide rate by developing novel interventions targeting risk and protective factors among Black youth. The overall objective of this application is to develop and validate new AI approaches to identify individual-level social risks of Black youth suicide and enhance trust within the underserved communities regarding the approaches of AI/ML.

In the United States, more than a third of patients are referred to a specialist each year, and specialist visits constitute more than half of outpatient visits. Even though all physicians highly value communication between primary care providers (PCPs) and specialists, both PCPs and specialists cite the lack of effective information transfer as one of the most significant problems in the referral process. Therefore, it is critical to investigate a new method to improve communication during care transitions. With their ubiquitous use, it is recognized that electronic health records (EHRs) should ensure a seamless flow of information across healthcare systems to improve the referral process. But, a lack of accessible and relevant information in the referral process remains a pressing problem. Recently, emerging deep learning (DL) and natural language processing (NLP) methods have been successfully applied in extracting pertinent information from EHRs and generating text summarization to improve care quality and patient outcomes. However, existing technologies cannot be applied to process heterogeneous data from EHRs and create high-quality clinical summaries for communicating a reason for referral. Responding to PA-20-185, this project will develop and validate a novel informatics framework to collect and synthesize longitudinal, multimodal EHR data for automatic referral form generation and summarization. While the referring provider and specialist can be any type of provider for any condition, the focus in this application has been on headache for primary care, because it is an extremely common symptom and affects people of all ages, races, and socioeconomic statuses. More importantly, relevant information needed for headache referrals has been defined in local and national evidence-based practice guidelines. Therefore, a health information technology solution to make these data accessible will empower communication between PCPs and specialists, which can improve the care of millions of patients suffering from disabling headache disorders. Based on our preliminary data and our experience with an interdisciplinary team of data scientists and physicians, we plan to execute specific aims: 1) Convert text-based guidelines into a standards-based algorithm for electronic implementation; 2) develop models to automatically populate data from EHR and clinical notes to fill the referral form; 3) create a framework to summarize the longitudinal clinical notes to fill out the referral form; and 4) develop and validate the headache referral system with a user-centered design approach. The research proposed in this project is novel and innovative because it will produce and rigorously test new solutions to improve the communication between health professionals to ensure that safe, high-quality care is provided and care continuity is maintained. The success of this project will (1) fill important gaps in our knowledge of understanding the types of information exchange that will optimize patient care during transitions and (2) provide evidence-based solutions to enable the exchange.

The recent COVID-19 pandemic has revealed the fragility of humankind. In our highly connected world, infectious disease can swiftly transform into worldwide epidemics. A plague can rewrite history and science can limit the damage. The significance of teamwork in science has been extensively studied in the science of science literature using transdisciplinary studies to analyze the mechanisms underlying broad scientific activities. How can scientific communities rapidly form teams to best respond to pandemic crises? Artificial intelligence (AI) models have been proposed to recommend scientific collaboration, especially for those with complementary knowledge or skills. But issues related to fairness in teaming, especially how to balance group fairness and individual fairness remain challenging. Thus, developing fair AI models for recommending teams is critical for an equal and inclusive working environment. Such a need could be pivotal in the next pandemic crisis. This project will develop a decision support system to strengthen the US-Australia public health response to infectious disease outbreak. The system will help to rapidly form global scientific teams with fair teaming solutions for infectious disease control, diagnosis, and treatment. The project will include participation of underrepresented groups (Indigenous Australians and Hispanic Americans) and will provide fair teaming solutions in broad working and recruiting scenarios.     This project aims to understand how scientific communities have responded to historical pandemic crises and how to best respond in the future to provide fair teaming solutions for new infectious disease crises. The project will develop a set of graph representation learning methods for fair teaming recommendation in crisis response through: 1) biomedical knowledge graph construction and learning, with novel models for emerging bio-entity extraction, relationship discovery, and fair graph representation learning for sensitive demographical attributes; 2) the recognition of fairness and the determinant of team success, with a subgraph contrastive learning-based prediction model for identifying core team units and considering trade-offs between fairness and team performance; and 3) learning to recommend fairly, with a measurement of graph-based maximum mean discrepancy, a meta learning method for fair graph representation learning, and a reinforcement learning-based search method for fair teaming recommendation. The project will support cross-disciplinary curriculum development by effectively bridging gaps in responsible AI and team science, fair project management, and risk management in science.

This collaborative project investigates the lack of diverse, representative datasets and insights in the development and use of technology. It explore the effects of disparities on the ability of technologists (e.g., practitioners, designers, software developers) to develop technology that addresses and mitigates systemic societal racism and historically marginalized individuals' ability to feel seen and heard in the technology with which they engage. The implications of this project are threefold: 1) it supports building relationships between technologists and technology users by understanding the values that most impact historically marginalized communities' engagement and data contributions; 2) given access to more diverse data and insights, the project provides technologists with interventions that empower them to make use of these data and insights in practice; 3) lastly, the work provides support and affirmation for the technologists who are already making these explicit considerations in their work without the adequate support. More broadly, insights from this project can be applied in practice to promote racial equity and ensure systemic racism is an explicit consideration in STEM education and workforce development by incorporating more equitable practices in technologists' workflow. This study seeks to answer three main research questions: 1) What are the barriers to engaging and amplifying marginalized voices in technological spaces and data sets for both technologists and users? 2) How can marginalized groups be engage when designing and developing data-centric systems without sacrificing their safety, security, and trust? 3) What does it look like to provide interventions for engaging the margins to technologists without compromising the safe spaces for marginalized groups? Using a multi-modal approach, the project will examine how researchers and technologists can best learn to engage in data-centric research with marginalized communities in an ethically and socially responsible manner that centers the rights and values of the communities of interest. Culturally relevant approaches and grounding philosophies will drive the research methods and analyses. Through surveys, semi-structured interviews, design workshops utilizing a combination of participatory design and community-based approaches, as well as case study analysis to collect qualitative and quantitative data, the research team will develop an intervention that supports technologists in responsible engagement. Aside from real-world implementation, this project will share its findings through academic and community-facing venues, such as journal publications, conference presentations, op-eds, blogs, workshops, and social media. This collaborative project is funded through the Racial Equity in STEM Education program (EDU Racial Equity). The program supports research and practice projects that investigate how considerations of racial equity factor into the improvement of science, technology, engineering, and mathematics (STEM) education and workforce. Awarded projects seek to center the voices, knowledge, and experiences of the individuals, communities, and institutions most impacted by systemic inequities within the STEM enterprise. This program aligns with NSF's core value of supporting outstanding researchers and innovative thinkers from across the Nation's diversity of demographic groups, regions, and types of organizations. Programs across EDU contribute funds to the Racial Equity program in recognition of the alignment of its projects with the collective research and development thrusts of the four divisions of the directorate.

The objective of this Research Project is to develop an advanced neural NLP modeling framework for interpretable and accurate text classification. Intuitively, when human users better understand model predictions (via model interpretability), the users can better use model predictions to augment their own human reasoning and decision-making. More generally, effective model explanations offer a variety of other potential benefits, such as promoting trust, adoption, auditing, and documentation of model decisions. Our modeling framework, ProtoType-based Explanations for Natural Language (ProtoTexNL), seeks to provide faithful explanations for model predictions in relation to training examples and features of the input text. 

Urban Air Mobility (UAM) envisions integrating the skyscape into the transportation network and encompasses services such as delivery drones, on-demand shared mobility by Vertical-Take Off and Landing (VTOL) aircraft for intra-city passenger trips, and, in the longer run, electric and autonomous VTOLs. This possible modal alternative provides a safe, reliable, and environmentally sound option to reduce surface-level congestion. Nevertheless, the history of transportation infrastructure development shows that it is imperative to design transportation infrastructures with the community to find the best balance between these sociotechnical requirements. Much research shows that the design of transportation systems has a long-lasting, often discriminatory effect that reinforces existing socio-economic inequality. As UAM is being developed as a new transportation mode, we are at an opportune moment to design its infrastructure to provide effective and equitable air mobility for all, avoiding our past mistakes. This project will focus on understanding the preferences, attitudes, and concerns of all stakeholders of UAM, including the potential users of UAM, the general public in different communities who may be positively and/or adversely affected by UAM, policymakers, and city planners. The knowledge elicited from the stakeholders will guide the design of an open-source Computer Aided Planning tool that policy-makers and urban planners can use to design UAM infrastructure that accommodates communities? priorities and enables transportation equity. While the timeline for UAM may be in the future, its deployment may entail significant future investment in infrastructure which makes inclusion of equity considerations and early community engagement critical. We propose a ''Community-in-the-Loop Integrative Framework for Fair and Equitable Urban Air Mobility (UAM) Infrastructure Design''. Our integrative framework will develop methods to engage with key stakeholders to address significant socio-technical challenges, including (a) understanding the community preferences and desiderata in terms of necessary considerations for equitable mobility, (b) developing novel machine learning techniques to generate design options that optimize for community desiderata efficiently and (c) devising community-driven evaluative measures and trade-off decision mechanisms. We address these challenges by drawing from urban and transportation engineering, aerospace, and computer and information sciences. The final product of our framework is an open-source Computer Aided Planning tool called VertiCAP. VertiCAP will be equipped with novel machine learning-based algorithms to navigate complex design space options, including long-term decisions (i.e., allocation of UAM airports, also known as vertiports), medium-term decisions (i.e., design of air space), and short-term decisions (i.e., air-traffic control). We will establish a ''community council'' representing different stakeholders. Through continuous interactions with the community council, we will evaluate and demonstrate the effectiveness of the developed VertiCAP tool in the City of Austin, TX and Southern California.

Algorithms are at the forefront of a transformative shift in the World of Work, profoundly influencing work dynamics, organizational structures, and the work environment. Despite their profound impact, a substantial knowledge gap exists concerning algorithmic management (AM) and its repercussions on occupational safety, health, and wellbeing. This gap is particularly pronounced in non-platform work settings, where AM's prevalence is growing. As the use of AM continues to expand across various economic sectors, it is imperative to investigate its effects on the wellbeing of workers. The overarching objective of the ALGOSH research program is to enhance our understanding of AM in non-platform sectors and its impact on the health, safety, and wellbeing of workers. Moreover, it aims to develop tools and strategies to mitigate associated risks. The three research aims of ALGOSH are: Facilitating the development of a standard for measurement of algorithmic management at work and related risks for health, safety and well-being. Increasing knowledge about the effects of algorithmic management on workers’ health, safety, and well-being. Investigating the balance of interests related to the control of algorithms in different legal contexts regarding occupational health and safety (OSH). To accomplish this mission, an international and interdisciplinary consortium of researchers has been assembled. For our research to have maximum societal impact, the program also has a strong stakeholder involvement and support from trade unions, business organizations, international bodies, and government agencies. Their collective efforts will examine, discuss, and assess the opportunities and challenges posed by algorithmic management, fostering a safer and healthier work environment for all. The program applies multiple methods including quantitative, qualitative, literature reviews and participatory research. 

Software systems have become an integral part of public and private sector management, assisting and automating critical human decisions such as selecting people and allocating resources. Emerging evidence suggests that software systems for algorithmic management can significantly undermine workforce well-being and may be poorly suited to fostering accountability to existing labor law. For example, warehouse workers are under serious physical and psychological stress due to task assignment and tracking without appropriate break times. On-demand ride drivers feel that automated evaluation is unfair and distrust the system?s opaque payment calculations which has led to worker lawsuits for wage underpayment. Shift workers suffer from unpredictable schedules that destabilize work-life balance and disrupt their ability to plan ahead. Meanwhile, there is not yet an established mechanism to regulate such software systems. For example, there is no expert consensus on how to apply concepts of fairness in software systems. Existing work laws have not kept pace with emerging forms of work, such as algorithmic management and digital labor platforms that introduce new risk to workers, including work-schedule volatility and employer surveillance of workers both on and off the job. To tackle these challenges, we aim to develop technical approaches that can (1) make software accountable to existing law, and (2) address the gaps in existing law by measuring the negative impacts of certain software use and behavior, so as to help stakeholders better mitigate those effects. In other words, we aim to make software accountable to law and policy, and leverage it to make software users (individuals and firms) accountable to the affected population and the public. This project is developing novel methods to enable standards and disclosure-based regulation in and through software systems drawing from formal methods, human-computer interaction, sociology, public policy, and law throughout the software development cycle. The work will focus on algorithmic work scheduling, which impacts shift workers who make up 25% of workers in the United States. It will take a participatory approach involving stakeholders, public policy and legal experts, governments, commercial software companies, as well as software users in firms and those affected by the software?s use, in the software design and evaluation. The research will take place in three thrusts in the context of algorithmic scheduling: (1) participatory formalization of regulatory software requirements, (2) scalable and interactive formal methods and automated reasoning for software guarantees and decision support, and (3) regulatory outcome evaluation and monitoring. By developing accountable scheduling software, the project has the potential for significant broader impacts by giving businesses the tools they need for compliance with and accountability to existing work scheduling regulations, as well as the capacity to provide more schedule stability and predictability in their business operations.

As part of the NIH Common Fund’s Bridge2AI program, the CM4AI data generation project seeks to map the spatiotemporal architecture of human cells and use these maps toward the grand challenge of interpretable genotype-phenotype learning. In genomics and precision medicine, machine learning models are often "black boxes," predicting phenotypes from genotypes without understanding the mechanisms by which such translation occurs. To address this deficiency, project will launch a coordinated effort involving three complementary mapping approaches – proteomic mass spectrometry, cellular imaging, and genetic perturbation via CRISPR/Cas9 – creating a library of large-scale maps of cellular structure/function across demographic and disease contexts. These data will broadly stimulate research and development in "visible" machine learning systems informed by multi-scale cell and tissue architecture. In addition to data and tools, this project will implement a standards data management approach based on FAIR access and software principles, with deep provenance and replication packages for representation of cell maps and their underlying datasets; initiate a research program in ethical AI, especially as it relates to how maps will be used in genomic medicine and model interpretation; and stimulate a diverse portfolio of training opportunities in the emerging field of biomachine learning.