Project Description

In collaboration with social-work scientists, we have addressed three different challenges faced by homeless youth in the United States: (i) preventing HIV among homeless youth through the development of robust end-to-end influence maximization algorithms, which can maximize the efficiency of Popular Opinion Leader (POL) interventions; (ii) mitigating opioid abuse among homeless youth by removing the bottlenecks in COR-12 rehabilitation programs through a combination of causal inference and robust prediction-driven optimization that incorporates fairness constraints; and (iii) suicide prevention among homeless youth by developing algorithms for robust graph covering problems with group fairness constraints, which can maximize the efficiency of Gatekeeper training interventions. In collaboration with homeless shelters, our research group was the first to deploy influence maximization based algorithms in the real-world, e.g., we ran a multi-year clinical trial showing the effectiveness of my influence maximization algorithms in minimizing risky sexual behaviors among homeless youth.

Robust End-to-End Influence Maximization for HIV Prevention: Popular Opinion Leader (POL) is a community-level HIV prevention intervention based on diffusion of innovation principles in which a cadre of trusted, well-liked key opinion leaders are trained to endorse safer sexual behaviors among peers. While the influence maximization problem provides a nice computational framework to find high-quality opinion leaders (or "seeds") which can maximize the diffusion of HIV preventative information inside a social network of homeless youth, most prior work makes several strong assumptions (such as accurate knowledge of social network structure and diffusion models) which rarely hold in real-world domains involving marginalized communities. Motivated by these real-world uncertainties, we have designed the first algorithms and models for tackling uncertainties in influence maximization, leading to a new era of influence maximization research.

Causal Inference + Optimization for Opioid Abuse Prevention: COR-12 rehabilitation programs are designed for people suffering from Opioid Use Disorders, but they are difficult to implement because of ad-hoc assessments (made by rehabilitation centers) about the cause of opioid use among individual patients. In collaboration with social-work scientists, we developed CORTA, a novel software agent which optimizes the delivery of opioid rehabilitation services to homeless youth. CORTA collects data about opioid usage behaviors of homeless youth and uses that data to train high dimensional causal inference models which can predict susceptibility of homeless youth to opioid addiction, and can uncover causative factors that lead to opioid abuse. Finally, using counterfactual estimates derived from these causal inference models, CORTA solves novel Integer Linear Program (ILP) formulations to determine the optimal assignment of homeless youth to the correct rehabilitation programs. CORTA’s ILP formulation finds such optimal assignments by minimizing the expected number of homeless youth suffering from opioid addiction, while respecting fairness and limited capacity constraints faced by rehabilitation centers.

Fair and Robust Graph Covering for Suicide Prevention: Gatekeeper training involving identifying and training key nodes in social networks as gatekeepers who can actively look out for suicidal tendencies among their immediate neighbors in the network. While these problems can be cast as robust graph covering problems, state-of-the-art graph covering algorithms result in biased node coverage, i.e., they tend to discriminate individuals (nodes) based on membership in traditionally marginalized groups. In collaboration with social-work scientists, we incorporated group fairness constraints in robust graph covering problems, and developed tractable approximation schemes to solve these problems on real-world networks, which allows us to find the optimal set of gatekeepers for suicide prevention among homeless youth.

Publications