Noisy heteroclinic networks and
sequential decision making
Yuri Bakhtin,
Georgia Tech
Abstract:
I will talk about sequential decision making models based on
diffusion
along heteroclinic networks of dynamical systems, i.e.,
multiple
saddle-type equilibrium points connected by heteroclinic orbits.
The
goal is to give a precise description of the asymptotic behavior
in
the limit of vanishing noise. In particular, I will interpret
exit
times for stochastic dynamics as decision making times and give
a
result on their asymptotic behavior. I will report on extensive
data
on decision making in no a priori bias setting obtained in a
psychology experiment (joint with Joshua Correll, University
of
Chicago), and compare the data with the theoretical results. I
will
also show that the same behavior of exit times appears
innonequilibrium models of statistical mechanics.