Design of Robust & Adaptable Systems
Methods
We devise intelligent systems capable of demonstrating competence and adaptability in complex uncertain environments and sensitive social contexts. Our proposed approaches present attractive scalability properties, which ensure that they can be deployed in the real-world. They also possess robustness properties that make them particularly suitable for socially sensitive settings.
Social Good Applications
Preserving Biodiversity via Robust Optimization
We are devising data-driven robust AI approaches to assist severely resource constrained NGOs dedicated to biodiversity preservation. Indeed, a very effective strategy to help preserve animals and their habitats consists in purchasing or easing parcels of land over time, thereby precluding them from being lost to human degradation. This work is supported through an NSF CMMI Operations Engineering 4 year $535k grant entitled "Preserving Biodiversity via Robust Optimization." For this research, we have partnered with Conservation Biologists at Panthera, at the Wildlife Conservation Society, and at the U.S. Geological Survey. We will pilot our proposed conservation plans in Central America to help protect the Jaguar.
Illustration of land transformation impact on the jaguar range in South and Central America
“Land transformation is the single most important cause of extinction, and current rates of land transformation eventually will drive many more species to extinction.” -- Science (Vitousek et al., 1997)
The gorilla, the elephant, and the jaguar, just some of the species that are critically endagered and that we seek to protect in our conservation work
Robust and Fair Suicide Prevention Interventions
For example, we are investigating the problem of selecting a subset of nodes (individuals) in a virtual or physical (social) network that can act as monitors capable of "watching-out" for their neighbors (friends) when the availability or performance of the chosen monitors is uncertain. Such problems arise for example in the context of "Gatekeeper Trainings" for suicide prevention on college campuses (in particular on those campuses that are under-resourced). Our proposed algorithm is being prepared to be piloted in dorms at the University of Denver by our collaborator there in the School of Social Work (IRB under review). This work is supported by a grant from the Army Research Laboratory Army Research Office (subaward from USC School of Social Work). We also recently received generous support from the Living to Love Another Day Foundation to help deploy our algorithm on the University of Southern California campus.
“Gatekeeper” training intervention for suicide prevention under uncertainty in the performance and availability of monitoring nodes
QPR Gatekeeper training logo (top) and University of Denver campus, where the algorithmic suicide prevention intervention will be piloted
Robust Optimization for Landslide Risk Management
We are investigating data-driven approaches to characterize uncertainty in robust AI. Notably, this work is motivated from the problem of spreading influence on uncertain social networks where limited information about social ties can be uncovered by strategically querying individuals about their friends. This work is supported through an NSF S&CC 3 year $2.4m grant entitled "Landslide Risk Management in Remote Communities: Integrating Geoscience, Data Science, and Social Science in Local Context." For this work, we have partnered with Behavioral/Social Scientists, Physical Scientists, and Economists at RAND Corporation, Geological and Environmental Scientists at the University of Oregon, and partners at the Sitka Sound Science Center to help mitigate the risk from Landslides in the town of Sitka, Alaska.
Redoubt Lake landslide, Sitka, AK [Photo Credit: Sitka Sound Science Center]
Related Papers
Exploring algorithmic fairness in robust graph covering problems
(*) A. Rahmattalabi, P. Vayanos, A. Fulginiti, E. Rice, B. Wilder, A. Yadav, M. Tambe
In Proceedings of the 33rd Conference on Neural Information Processing Systems, NeurIPS, 2019.
note: acceptance rate ~21% in year of submission
(**) S.M. Mc Carthy, P. Vayanos and M. Tambe
In Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence (IJCAI), pp. 3770-3776, 2017.
Chance-constrained optimization for refinery blend planning under uncertainty
Y. Yang, P. Vayanos, and P.I. Barton
Industrial & Engineering Chemistry Research, 56 (42), pp. 12139–12150, 2017.
A constraint sampling approach for multi-stage robust optimization
P. Vayanos, D. Kuhn and B. Rustem
Automatica, 48(3):459-471, 2012.
Decision rules for information discovery in multi-stage stochastic programming
P. Vayanos, D. Kuhn, and B. Rustem
In Proceedings of the 50th IEEE Conference on Decision and Control, pp. 7368-7373, 2011.
Hedging electricity swing options in incomplete markets
P. Vayanos, W. Wiesemann, and D. Kuhn
In Proceedings of the 18th IFAC World Congress, pp.846-853, 2011.
Partner Organizations
Related Grants
OE: Preserving Biodiversity via Robust Optimization
National Science Foundation, Operations Engineering
Role: PI (with Co-PI: B. Dilkina)
Award #: OE-1763108
Total Award Period Covered: 07/15/2018-07/14/2022 (4 Years)
Total Award Amount: $535,335
Own Share: $403,638
S&CC: Landslide Risk Management in Remote Communities: Integrating Geoscience, Data Science, and Social Science in Local Context
Role: Co-PI (PI: Robert Lempert; Collaborative grant with RAND, University of Oregon, Sitka Sound Science Center)
Total Award Period Covered: 9/01/19-08/31/22
Total Award Amount: $2,100,974
Own Share: $216,218
Predictive Modeling for Early Identification of Suicidal Thinking in Social Networks
U.S. Army Research Laboratory
Role: PI on satellite from USC School of Social Work (PI from SW: Eric Rice)
Award ID: W911NF-17-1-0445
Satellite Award Amount: $12,420
Gift in Support of Center for AI in Society
Living to Love Another Day Foundation
Total gift amount: $5,000