No subject
Lina Massone
lina at ai.mit.edu
Wed Jun 20 17:54:51 EDT 1990
********** DO NOT FORWARD TO OTHER BBOARDS ***********
The following technical report is available.
Target-Switching Experiments
with a Sequential Neuro-Controller
Lina Massone
Dept. of Brain and Cognitive Sciences
Massachusetts Institute of Technology
77 Massachusetts Avenue - Cambridge Ma 02139
This paper describes some target-switching experiments simulated
with a neural system that drives a three-joint redundant limb.
The system is composed of a controller (a sequential network)
and a limb emulator. The system was trained to generate aiming
movements of the limb towards targets specified as sensory stimuli;
it was not trained to perform the target-switching task itself.
The experiments demonstrate that the system possesses the ability
to solve the target-switching task, which requires generalization
with respect to both initial limb posture and sensory stimulation.
I performed the experiments under two different perceptual conditions:
(1) on/off switching of the two stimuli, (2) temporal overlap of the
two stimuli. The second case refers to a hypothesis proposed
by many experimental investigators about two different systems
being involved in the programming of movements: a "where-system"
that would build an internal representation of the target that
shifts gradually to the new values, and a "when-system" that
would start the motor program generator.The "where-system"
would be able to account for the observed differences in path, while the
"when-system" would be able to account for the response-time phenomenon.
The case of temporal overlap of the two stimuli is a simulation
of the "where-system". I present a qualitative comparison of data
generated by the neural system under conditions (1) and (2),
namely (i) the endpoint paths and velocity profiles,
(ii) the patterns of muscular activation. Results of the
comparison show that in presence of the "where-system" the controller
can account for the variability in paths and the basic two-peak
structure of the velocity profiles commonly observed in psychophysical
experiments. In absence of the "where-system" the behavior of the
controller is, on the contrary, highly stereotyped. Results also
point out an inadequacy in the network architecture to deal with
the observed high peak velocities after stimuli are switched.
Please forward all requests to lina at ai.mit.edu
More information about the Connectionists
mailing list