At my research group, we conduct research in Artificial Intelligence (AI) to solve some of the most intractable social problems facing the world today. Specifically, we work on advancing AI research motivated by the grand challenges of the American Academy of Social Work and Social Welfare and the UN Sustainable Development agenda, which positively impact the lives of under-served communities in society. As a result, we collaborate actively with practitioners and scientists in various disciplines, including computer science, social work, operations research, psychology, economics, criminology, conservation biology and ecology.
Examples of research projects include (i) AI for raising awareness about HIV among homeless youth; (ii) AI for promoting mental health awareness among PTSD veterans; (iii) AI for convincing people to stop using drugs by using Peer-Influence effects in social networks; and (iv) AI for promoting healthier eating habits among children in public schools.
Our use-inspired research (i.e., drawing motivation from real-world societal applications) relies heavily on the fields of machine learning, multi-agent systems and reasoning under uncertainty, social network analysis, computational game theory, mechanism design and convex/combinatorial optimization.
This project focuses on the study of diffusion processes in social networks of hard to reach populations (such as homeless youth) in order to spread information and raise general levels of awareness about dangerous diseases (such as HIV) among such populations. On a humanitarian level, the end goal of this project is to reduce rates of HIV infection among disadvantaged populations by influencing and inducing behavior change in homeless youth populations that drives them towards safer practices, such as regular HIV testing, etc. On a scientific level, the goal is not only to model these influence spread phenomena, but to also develop decision support systems (and the necessary tools/algorithms/mechanisms) using which information can be spread in the social networks of homeless youth in the most efficient manner. Our primary focus in this project is to develop algorithms and tools which are actually usable and deployable in the real world, i.e., algorithms which can actually benefit society for good. In fact, we strive to validate all our models, algorithms and techniques in the real world by testing it out with actual homeless youth (specifically youth in Los Angeles). Over the past three years, we have been collaborating with social workers from Safe Place for Youth (SPY) and My Friend’s Place (homeless shelters in Los Angeles) to understand the problems that they face in raising awareness about HIV (and other STDs) among homeless youth, come up with innovative ways to solve their problems, and finally test out our algorithms by doing pilot deployment studies with actual homeless youth.
Project Webpage: HTML
Security is a critical concern around the world, whether it is the challenge of protecting ports, airports and other critical national infrastructure, or protecting wildlife/forests and fisheries, or suppressing crime in urban areas. In many of these cases, limited security resources prevent full security coverage at all times. Instead, these limited resources must be allocated and scheduled efficiently, avoiding predictability, while simultaneously taking into account an adversary’s response to the security coverage, the adversary’s preferences and potential uncertainty over such preferences and capabilities. Computational game theory can help us build decision-aids for such efficient security resource allocation problems.
A key deficiency in current social network research (including influence maximization problems) is that nodes in the social networks are represented as abstract entities. In this work, we make a departure from this literature by considering nodes of the social network (i.e., human beings) as self-interested agents. For example, in domains such as poverty alleviation and environment sustainability, humans (in the social network) have their own personal incentives which need to be satisfied in order for them to get influenced (and for them to spread influence). We aim to answer basic questions including how to model game theory and influence maximization in an integrated manner, defining appropriate equilibrium solution concepts, and incentivization mechanisms to achieve these notions of equilibrium. Moreover, introducing game theoretic aspects into influence maximization would require tackling a multitude of fundamental research challenges such as uncertainties about game and model parameters, learning accurate human behavior models to find optimal game theoretic strategies.
Most previous work in influence maximization assumes that influence spreads in the network in discrete time steps, with no regards to the spatial and temporal factors that may hinder or facilitate influence spread. However, these assumptions are not adequate to model real-world social phenomena. Moreover, in many real-world domains, the nodes in a social network (or the influencers) act in a geographical space over time. Therefore, it is important to develop models and algorithms which tackle spatio-temporal aspects such as continuity of the influence spread process over space and time, complex spatial constraints and dynamic behavior patterns (that limit possible paths of influence spread).
The field of influence maximization has made rapid advances, resulting in many sophisticated algorithms for identifying “influential” members in social networks. However, in order to engender trust in influence maximization algorithms, the rationale behind their choice of “influential” nodes needs to be explained to its end-users. This is a challenging open problem that needs to be solved before these algorithms can be successfully deployed on a large scale. This project attempts to tackle this open problem via four major contributions: (i) we propose a machine learning based paradigm for designing explanation systems for influence maximization algorithms by exploiting the trade-off between an explanation’s accuracy (or correctness) and its interpretability; our novel paradigm treats influence maximization algorithms as black boxes, and is flexible enough to be used with any such algorithm; (ii) we utilize this paradigm to build XplainIM, a suite of explanation systems which can explain the solutions of any influence maximization algorithm; (iii) we illustrate the usability of XplainIM by using it to explain the solutions of a recent influence maximization algorithm to ∼200 human subjects on Amazon Mechanical Turk (AMT); and (iv) we provide extensive analysis of our AMT results, which shows the effectiveness of XplainIM in explaining solutions of influence maximization algorithms.
A large body of work on efficient algorithms with theoretically provable guarantees to solve the influence maximization problem has been developed over the last two decades. Unfortunately, none of these algorithms have been tested in the real-world (i.e., physical real-world, not virtual), which brings into question their real-world usability. Moreover, some of these models have perhaps been chosen for their mathematical elegance, but their ability to closely mirror diffusion spread in the real-world has not been validated. By analyzing data collected from real-world deployments of these diffusion propagation models, we have proposed novel diffusion models which significantly outperform the widely accepted diffusion models (e.g., independent cascade, linear threshold) in predicting real-world diffusion patterns.