Amy Sliva

My general research interests are in the field of computational social science, at the intersection of artificial intelligence (AI) and political science, sociology, and psychology. Specifically, I work in AI and reasoning under uncertainty, using probabilistic models to understand, analyze, and forecast complex sociocultural problems that often consist of several interacting and interdependent systems along the PMESII--political, military, economic, social, information, and infrastructure--dimensions. I have developed several scalable frameworks for sociocultural modeling that leverages both qualitative social science theories and quantitative computational methods--adapting techniques from probabilistic graphical models, formal logics, systemic functional linguistics, and machine learning--in a variety of application domains, including food security, intergroup conflict, and cybersecurity. Such modeling approaches aim to support decision makers to explore and identify possible mitigating policies.

I am also interested more broadly in issues of technology policy and the social and security impact of information and communcation technology, particularly regarding governance and cybersecurity.

Some of my past and current research projects include:

• Decision Analysis for Food Security--Novel combination of probabilistic graphical modeling and sensitivity analysis techniques to quantify uncertainty and resiliencing for decision makers, applied to a case study of food security issues in Gambella, Ethiopia. This work was originally funded through the DARPA World Modeler's program and involved collaboration with the Bill an Melinda Gates Foundation.
• Fake News Detection--Applicaiton of machine learning techniques to identify "fake news" articles based on propagation patterns rooted in social psychology theories.
• Cyber Attack Prediction--Ensemble modeling framework for predicting cyber attacks based on open source "unconventional" sensors, such as social media posts, dark web chatter, and economic indicators, leveraging a variety of AI modeling approaches, including logistic regression predictors, sociolinguistic models, and social science theories.
• Decision Making Under Threat--Models of how media amplificaiton of violent events impacts public threat (mis)perception and the policy decisions of leaders that may cause persistent conflict.
• Cyber attacker behavioral models--Probabilistic models for forecasting attacker intentions in cyber attacks, facilitating better decision making regarding defensive resource allocation and potential counterintelligence activities.
• Cybersecurity Framework--Collaboration with the National Defense University Institute for National Strategic Studies to investigate a conceptual and strategic framework for cyberwar, focusing on the real and bureaucratic delineations between crime, espionage, and war.
• Actionable State Change Attempts--Framework for modeling the optimal response policy or counterstrategy to an agent's behavior, balancing the cost with likelihood of success.
• Technology Integration in the U.S. Intelligence Community--Investigating the technical and institutional barriers to technology integration in the Intelligence Community and assessing the needs for adaptive technologies.
• Stochastic Opponent Modeling Agents (SOMA)--Probabilistic logic modeling and reasoning framework for cultural dynamics. SOMA has been used primarily to model the behaviors of terror organizations, such as Hezbollah, Hamas, and Lashkar-e-Taiba, as well as the key players in the Afghan drug economy.
• SOMA Terror Organization Portal (STOP)--Online tool that provides national security experts, policy analysts, and political science researchers with access to data on terror organizations and the behavioral modeling and forecasting tools developed by LCCD.
• Change Analysis Predictive Engine (CAPE)--Probabilistic framework for modeling and forecasting when and how a group may change its behavior or strategies.