Jonathan Cannon, head of the Trimba Lab at McMaster University will give a CTN Seminar on Dynamic inference in rhythm perception, production, and synchronization in E5 2004 Sept 26 at 15:30.
Abstract:
Moving
in
time
with
rhythmic
music
is
nearly
universal
across
human
cultures,
and
group
rhythmic
coordination
produces
remarkable
group
cohesion
effects,
in
part
by
dissolving
subjective
boundaries
between
self
and
other.
How
can
we
make
sense
of
this
unique
sensorimotor
behavior
in
the
context
of
the
wide
human
repertoire
of
perceptual
and
motor
processes?
In
this
talk,
I
propose
that
the
theory
of
Bayesian
predictive
processing
provides
not
only
a
conceptual
framework
but
also
a
clear,
intuitive
mathematical
modeling
language
for
rhythm
perception,
rhythm
production,
and
sensorimotor
synchronization
through
self/other
integration.
Following the predictive processing account of perceptual inference, I propose a computational model in which we perform approximate Bayesian inference to estimate the momentary phase and tempo of ongoing underlying metrical cycles using learned metrical models. Then, drawing on the theory of “active inference” which extends predictive processing to the realm of action, I propose a closely related computational model of rhythm production as a closed loop: timely feedback from our actions informs a dynamic model of our moving body, and that model guides the timing of subsequent action. Bringing these two computational models together, I propose a new formal account of sensorimotor synchronization: by modeling a heard rhythm and our own motor feedback as though they arise from the same underlying metrical cycle (i.e., modeling “self” and “other” as a single unified process), active inference naturally brings our actions into synch with what we hear. I explore evidence for this model, its new predictions, and experiments that might test those predictions.