Research

Allston campus

New home of Teamcore group on Allston campus Current Projects Public Health, Conservation, Safety and Security from HIV and tuberculosis prevention to wildlife conservation...

Research

We have organized our projects into three large domain areas: 1) social work, public health, and medical decision making, 2) conservation, and 3) safety and security. Our recent research on the combating the COVID-19 pandemic is also listed above as a fourth category. Please click on the icons above to view more specific projects. 

We focus on fundamental research problems in computational game theory, machine learning, automated planning, intelligent agents and multiagent interactions driven by real world challenges, ensuring a virtuous cycle of research and real-world applications. Examples of research driven by real-world problems include algorithms that enable influence maximization in social networks under dynamism and uncertainty, challenges of learning and handling bounded rationality models of human (adversary) behavior from real-world data, solving massive scale games uncertainty, and others.

Our main research accomplishments include the following:

Public safety and security

Pioneering and foundational research in Security games for public safety: We provided the first ever applications of computational game theory for operational security, starting with deployment of game theoretic algorithms for security (e.g., counter terrorism) at LAX airport in 2007. This is one of the largest and busiest airports in the United States that serves 80 Million passengers. This research was followed by pioneering deployments of security games used by major security agencies for the allocation of limited security resources, including the US Coast Guard, the US Federal Air Marshals and others. This research mentioned in testimonies given in US congressional subcommittee hearings on three separate occasions – by officials of the LAX police, the Federal Air Marshals Service, and the US Coast Guard -- as having led to improvements due to AI/game theory for public safety (in 2008, 2012 and 2013).

The key challenge is to optimize the limited security resources for public safety. Our key insight was to model the problem as a security game, and provide models and efficient algorithms for obtaining such optimal security allocation policies. As a result of this work, we have been awarded the US Coast Guard Meritorious Team Commendation from the Commandant of the Coast Guard (2013), Certificate of Appreciation from the US Federal Air Marshals Service (2011) and special commendation given by the Los Angeles World Airports police from the city of Los Angeles (2009).

Wildlife Conservation

First large-scale applications of machine learning and game theory for wildlife conservation: Based on our green security game model and algorithms, which combine machine learning with security games, we are the first to apply AI models for global anti-poaching project. Predicting snare locations using machine learning enables rangers to remove them more effectively. Conducted over several years in cooperation with NGOs, this work has had substantial real-world impact resulting in removing hundreds of snares set to trap endangered animals, in Uganda and Cambodia. This work also led to a 5-fold increase in snare removals in Cambodia. The SMART partnership – a collaboration of major conservation agencies including WCS and WWF -- has begun deploying our algorithms in 800 wildlife parks internationally, bringing AI to the fight to save the lives of endangered animals all around the globe.

Public Health

Pioneering research for first large scale applications of social network algorithms for public health: We pioneered practical applications of algorithms on social networks for positive social interventions in low-resource communities. We conducted a large-scale field study to spread information about HIV prevention among youth experiencing homelessness (YEH), a population that has 10 times the rates of HIV than the housed population. Empirical results with 750 YEH in in Los Angeles demonstrated that our AI-guided interventions were far more effective than traditional methods in spreading HIV prevention information. 

Early Research in Foundational models of multiagent systems

Ours was one of the first papers which operationalized a comprehensive architecture for teamwork among intelligent agents that was founded on formal BDI-logic theories of teamwork, allowing “team-oriented programming” –facilitating programming of teams of agents. This paper: “M. Tambe. Towards flexible teamwork. Journal of Artificial Intelligence Research (JAIR), 1997.” won the “Influential paper award” of the International Foundation for Agents and Multiagent Systems at AAMAS 2012. We showed the effectiveness of this teamwork architecture in making it easier to build agent teams in many different domains.

We contributed with our collaborators the foundational paper in Distributed Constraint Optimization (DCOP) that started that subfield by providing the first comprehensive algorithm with quality guarantees (see P. Modi, W. Shen, M. Tambe, M. Yokoo “ADOPT:Asynchronous distributed constraint optimization with quality gurantees” Artificial Intelligence Journal. 2005) A whole variety of research groups picked up on DCOP research and this remains an area under continued investigations.

Our earlier paper introducing a novel Distributed POMDP model for teamwork on “The communicative multiagent team decision problem” won the AAMAS’02 Best paper. This and subsequent papers on dec-POMDPs led to a significant number of follow-on papers by others.

Past Projects