|
|
|
Opening conference on:
“Motor adaptation
and the timescales of memory”
Wednesday June 6th,
2007 from 2:00 to 3:15pm
Location: Auditorium
Euler, UCL Louvain-la-Neuve
|
|
|
|
Detailed program:
Wed-Thu-Friday June 6-7-8
from 2:00 to 6:00pm
Location: Auditorium
Euler, UCL Louvain-la-Neuve
Wednesday
- Motor Adaptation and the Timescales of Memory
(1:15)
As our brain generates motor commands, it also predicts the sensory
consequences. Why is prediction
a fundamental part of motor control?
One can provide two reasons: first, the ability to predict
allows us to sense the world better than is possible from our sensors
alone. Second, the ability to
predict overcomes the fundamental limitation of time delay in our
sensory system. However, for
prediction to be valuable, it has to be accurate, which implies that
adaptation is also an integral part of the brain mechanisms that make
predictions. In the first part
of this lecture, I will focus on saccade adaptation and demonstrate
that the motor memory that supports learning is composed of multiple
timescales: a fast system that strongly responds to error but rapidly forgets, and a slow system that weakly responds to
error but has excellent retention.
The simple model appears to be able to account for a wide body
of data, including learning in reaching and certain aspects of
declarative memory.
In the second part of the lecture I ask the question of why the
nervous system should learn motor control in this way. Why should we have an adaptive
system that rapidly forgets? I
argue that there is a link between how our motor system learns and the
natural events that can affect the motor system: events like fatigue,
aging, and disease. That is, we
forget as a function of time because certain perturbations (like
fatigue) naturally go away as a function of time. Once again I focus on the saccadic
system and show that the timescales of fatigue are surprisingly
similar to timescales of adaptation to perturbations. The mathematical framework that we
will use to model saccades and other kinds of movements like reaching
is optimal control.
Vaziri S, Diedrichsen
J, and Shadmehr R (2006) Why does the brain predict sensory
consequences of oculomotor commands? Optimal integration of the
predicted and the actual sensory feedback. Journal of
Neuroscience, 26:4188-4197. Paper
Kording KP, Tenenbaum
JB, and Shadmehr R (2007) The dynamics of memory as a consequence
of optimal adaptation to a changing body. Nature
Neuroscience, 10:779-786. Paper
Smith MA, Ghazizadeh A, and Shadmehr R (2006) Interacting adaptive processes with multiple timescales underlie
short-term motor learning. Public Library of Science
Biology, 4:e179. Paper
Synopsis
- A computational view of motor control (0:50)
This lecture introduces the problem of motor control from a
computational perspective. The
act of making a movement involves solving four kinds of problems: 1) We need
to learn the costs that are associated with our actions as well as the
rewards that we may experience upon completion of that action. 2) We need to learn how our motor
commands produce changes in state of our body and our environment. 3) Given the cost structure of the
task and the expected outcome of motor commands, we need to find those
motor commands that minimize the costs and maximize the rewards. 4) Finally, as we execute the motor
commands, we need to integrate our predictions about sensory outcomes
with the actual feedback from our sensors to update our belief about
our state. In this framework,
the function of basal ganglia appears related to learning costs and
rewards associated with our sensory states. The function of the cerebellum is to
predict sensory outcome of motor commands and correct motor commands
through internal feedback.
Together, reward driven optimal feedback control theory appears
the most consistent framework to explain a number of disorders in
human motor control.
Shadmehr R (2007) A computational view of
motor control. In: The Encyclopedia of Neuroscience. Squire LR (editor), Elsevier, in
press. Paper
Thursday
- The problem of state prediction: mathematical
background (1:15)
This lecture introduces the mathematics of optimal state estimation,
focusing on linear systems and the Kalman
filter. Topics include optimal
parameter estimation, parameter uncertainty, state noise and
measurement noise, adjusting learning rates to minimize model
uncertainty. Derivation of the Kalman filter algorithm.
- The problem of state prediction: classical
conditioning, integration of predictions with observations, and the
multiple timescales of memory (1:15)
This lecture applies the optimal estimation algorithm to biological data:
classical conditioning in animals, data fusion and combining data from
multiple sensors, fast and slow memory systems, massed vs. spaced
learning, forward models and integration of predicted with measured
sensory outcomes.
Dayan P, and Yu AJ (2003) Uncertainty and learning. IETE Journal of Research 49:171-182. Paper
Kording KP, Tenenbaum
JB, and Shadmehr R (2007) The dynamics of memory as a consequence
of optimal adaptation to a changing body. Nature
Neuroscience, 10:779-786. Paper
Friday
- Optimal control theory: introduction (1:15)
This lecture introduces optimal control using the method of Lagrange
multipliers, focusing on open-loop optimal control.
Todorov E (2004) Optimality principles in
sensorimotor control. Nature Neuroscience Reviews
7:907-915.
- Optimal stochastic feedback control: a framework
for biological motor control (1:15)
This lecture introduces feedback into the optimal control problem,
focusing on stochastic optimal feedback control using Gaussian and
signal dependent noise.
Todorov E (2005) Stochastic optimal control and
estimation methods adapted to the noise characteristics of the sensorimotor
system. Neural Computation
17:1084-1108
|
|
|
|
|
last update: May 24th, 2007
Author: Philippe LEFEVRE
|