Title: Behavior Model Calibration for Epidemic Simulations
Speaker: Meghendra Singh
Abstract: Large-scale simulations of human populations are a powerful tool for informed policy making. With respect to computational epidemiology, such simulations are used to study disease outbreaks and assess intervention strategies. A recent research direction in this area is the study of human behavioral factors in determining epidemic outcomes. In this talk I will present an approach to calibrate a behavior model derived from an epidemiological survey using a large-scale flu simulation with Episimdemics. We use an inverse reinforcement learning approach, where we have to learn the cost function, for modeling a simulated agent’s decision-making specific to avoidance behaviors. We use gradient descent to search for optimal costs so that simulated behaviors match the survey data associated with these behaviors, using the flu simulation in the optimization loop. This results in a calibrated decision making model for simulated agents that is consistent with the epidemiological survey responses.
Title: Inferring Local Transition Functions of Discrete Dynamical Systems From Observations of System Behavior
Speaker: S. S. Ravi
Abstract: We consider the problem of inferring the local transition functions of discrete dynamical systems from observed behavior. Our focus is on synchronous systems whose local transition functions are threshold functions. We assume that the topology of the system is known and that the goal is to infer a threshold value for each node so that the system produces the observed behavior. We show that some of these inference problems are efficiently solvable while others are NP-complete, even when the underlying graph of the dynamical system is a simple path. We identify a fixed parameter tractable problem in this context. We also consider constrained versions of threshold inference problems where the input includes a set of equality or inequality constraints (which specify pairs of nodes which must have the same threshold value or different threshold values). This work was done jointly with Abhijin Adiga, Chris Kuhlman, Madhav Marathe (NDSSL), Daniel Rosenkrantz and Richard Stearns (University at Albany -- SUNY).
Meeting number: 643 164 789
Host key: 749673
+1-855-749-4750 US TOLL FREE
+1-415-655-0001 US TOLL
Show toll-free dialing restrictions
Access code: 643 164 789