Speaker
Linh Huynh
(Dartmouth College)
Description
Stochastic switching systems arise in a wide range of applications in which agents adapt to dynamically changing environments. Two seemingly distinct yet analogous applications are cancer cells adapting to drug environments and large language models adapting to human interaction environments. In this work, I present a collective active inference framework for such systems, in which agents make optimal decisions by quantifying uncertainty from partially observable data. This framework is related to sampling from the stationary distribution of a spin glass model known as a Boltzmann machine. More broadly, this work establishes a new direction for applying spin glass theory to cancer evolution and artificial intelligence.
Author
Linh Huynh
(Dartmouth College)
