Research

Research Interest

Artificial Intelligence, Computational Game-theory, Mechanism Design, Machine Learning, Spatio-temporal Analysis, Large-scale and Robust Optimization, Security and Sustainability

Overview of Research

My research interests lie in the area of artificial intelligence, focusing on the intersection of computational game theory and machine learning, with strong connections to robust and large-scale optimization. Aiming to address real-world challenges in critical domains such as security, sustainability, and mobility, my work not only has strong theoretical contributions but also has led to applications that have fundamentally altered current practices in the domains I have worked in. Along the way, I have actively collaborated with researchers and practitioners in diverse disciplines, including computer science, operations research, psychology, economics, criminology, conservation biology and ecology.

My work has been featured in media including:

Projects

Pricing Mechanism for Ridesharing Platforms

Myopic pricing mechanism for ridesharing platforms may lead to drivers strategically rejecting ride requests (e.g., cancel the ride after calling you asking your destination). We work on designing pricing mechanisms that maximize total values of passengers picked up and align incentives of drivers.

  • Spatio-Temporal Pricing for Ridesharing Platforms
    Hongyao Ma, Fei Fang, David C. Parkes
    [ArXiv version]

Detect Illegal Mining Sites from Satellite Imagery

We built a tool identifying mining sites from satellite imagery. Code package is available here [GitHub Repository]

This slideshow requires JavaScript.

Detecting Poachers and Wildlife from Conservation Drones

The unrelenting threat of poaching has led to increased development of new technologies to combat it. One such example is the use of long wave thermal infrared cameras mounted on unmanned aerial vehicles (UAVs or drones) to spot poachers at night and report them to park rangers before they are able to harm animals. However, monitoring the live video stream from these conservation UAVs all night is an arduous task. Therefore, we build SPOT (Systematic POacher deTector), a novel application that augments conservation drones with the ability to automatically detect poachers and animals in near real time. The promising results from the test in the field have led to a plan for larger-scale deployment in a national park in Botswana. While SPOT is developed for conservation drones, its design and novel techniques have wider application for automated detection from UAV videos.

 

  • SPOT Poachers in Action: Augmenting Conservation Drones with Automatic Detection in Near Real Time
    Elizabeth Bondi, Fei Fang, Mark Hamilton, Debarun Kar, Donnabell Dmello, Jongmoo Choi, Robert Hannaford, Arvind Iyer, Lucas Joppa, Milind Tambe, Ram Nevatia
    In IAAI-18: The Thirtieth Conference on Innovative Applications of Artificial Intelligence, February 2018
    [pdf]

Protect Wildlife from Poaching

Poaching of endangered species is reaching critical levels as the populations of these species plummet to unsustainable numbers. The global tiger population, for example, has dropped over 95% from the start of the 1900s and has resulted in three out of nine species extinctions. Depending on the area and animals poached, motivations for poaching range from profit to sustenance, with the former being more common when profitable species such as tigers, elephants, and rhinos are the targets.

To counter poaching efforts and to rebuild the species’ populations, countries have set up protected wildlife reserves and conservation agencies tasked with defending these large reserves. Because of the size of the reserves and the common lack of law enforcement resources, conservation agencies are at a significant disadvantage when it comes to deterring and capturing poachers. Agencies use patrolling as a primary method of securing the park. Due to their limited resources, however, patrol managers must carefully create patrols that account for many different variables (e.g., limited patrol units to send out, multiple locations that poachers can attack at varying distances to the outpost).

PAWS takes basic information about the protected area and information about previous patrolling and poaching activities as input, and generates patrol routes as output. As the patrollers execute the patrol routes, more poaching data will be collected, and feed back to PAWS. The core algorithm of PAWS integrates learning poachers’ behavior model, game-theoretic reasoning and route planning. More specifically, PAWS learns the behavior models of the poachers from the crime data collected. Based on the poachers’ behavior model, PAWS calculates a randomized patrolling strategy, in the form of a set of patrol routes and the probabilities of taking each route. PAWS then suggests patrol routes sampled from this strategy to the patrollers.

A preliminary field test of PAWS was conducted in Uganda’s Queen Elizabeth National Park (QENP) in April 2014. PAWS patrols were outputted onto a GPS unit as a series of waypoints. Using the set of waypoints on the GPS as a directional guide, wildlife rangers executed their patrol and searched for signs of illegal activity.

  • Taking it for a Test Drive: A Hybrid Spatio-temporal Model for Wildlife Poaching Prediction Evaluated through a Controlled Field Test
    Shahrzad Gholami, Benjamin Ford, Fei Fang, Andrew Plumptre, Milind Tambe, Margaret Driciru, Fred Wanyama, Aggrey Rwetsiba, Mustapha Nsubaga, Joshua Mabonga
    In ECML-PKDD 2017: The European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases
    [pdf|ECML version]
  • Deploying PAWS: Field Optimization of the Protection Assistant for Wildlife Security
    Fei Fang, Thanh H. Nguyen, Rob Pickles, Wai Y. Lam, Gopalasamy R. Clements, Bo An, Amandeep Singh, Milind Tambe, Andrew Lemieux
    In IAAI-16: The Twenty-Eighth Annual Conference on Innovative Applications of Artificial Intelligence, February 2016
    Winner of Innovative Application Award
    [pdf|IAAI version|bib]

[Check this project at CMU – Societal Computing Program website]

Patrol Strategies for Protecting Ferries

ferryescort

This project focuses on designing patrol strategies to protect the ferries that are moving between terminals with fixed schedules. Packed with hundreds of passengers, these may present attractive targets to attack (e.g., with a small boat packed with explosives that may be only detected once it gets close to the ferry). Small, fast patrol boats can provide protection to such ferries, but there are often limited numbers of patrol boats, i.e., they cannot protect the ferries at all times at all locations.

I propose a new game model for this problem with a discretized strategy space for the defender and a continuous strategy space for the attacker, and provided an efficient linear program-based solution that uses a compact representation for the defender’s mixed strategy, while accurately modeling the attacker’s continuous strategy using a novel sub-interval analysis method. I also propose heuristic methods of equilibrium refinement for improved robustness. This work has been deployed by the US Coast Guard for protecting the Staten Island Ferry in New York City since 2013 and fundamentally altering previously used tactics.

  • Protecting Moving Targets with Multiple Mobile Resources
    Fei Fang, Albert X. Jiang, Milind Tambe
    In JAIR: Journal of Artificial Intelligence Research, 48:583-634, 2013
    [pdf|JAIR version|bib]

[Check this project at CMU – Societal Computing Program website]

Patrol Strategies for Protecting Forest Land

forestpatrol

Illegal extraction of fuelwood or other natural resources from forests is a problem confronted by officials in
many developing countries. We focus on finding the best patrol strategy which distributes the patrols throughout the forest,
in space and time, in order to minimize the resulting amount of extraction that occurs or maximize the degree of forest protection.

We build a game model for this problem where the actions of adversary are taken over continuous space (due to the continous forest land) and provide an algorithm computing the optimal distribution of patrol effort.

  • Policy Learning for Continuous Space Security Games using Neural Networks
    Nitin Kamra, Umang Gupta, Fei Fang, Yan Liu, Milind Tambe
    In AAAI-18: The Thirty-Second AAAI Conference on Artificial Intelligence, February 2018
    [pdf]
  • Patrol Strategies to Maximize Pristine Forest Area
    Matthew P. Johnson, Fei Fang, Milind Tambe, Heidi Jo Albers
    In AAAI-12: The Twenty-Sixth Conference on Artificial Intelligence (Computational Sustainability Track), July 2012
    [pdf|AAAI version|bib]

Computational Game Theory for Protecting Fisheries

fish

Illegal, unreported, and unregulated (IUU) fishing is one of the major threats to the sustainability of ocean fish resources. It is impossible to maintain a 24/7 presence to prevent IUU fishing everywhere due to the limited asset patrolling resources. Hence the allocation of the patrolling resources becomes a key challenge for security agencies like USCG. Research within fishery protection project aims to address the problem of deriving accurate patrol schedules for the US Coast Guard. [Check the project website]

In our work, we focus on robust protection of fisheries against unknown a population of heterogeneous Lanchas of unknown type who fish frequently, and we have very little data about these Lanchas in general.

  • Robust protection of fisheries with COmPASS
    William Haskell, Debarun Kar, Fei Fang, Milind Tambe, Sam Cheung, Elizabeth Denicola
    In IAAI-14: The Twenty-Sixth Annual Conference on Innovative Applications of Artificial Intelligence, July 2014
    [pdf|IAAI version|bib]