Robustness of VOR and OKR adaptation under kinematics and dynamics transformations

Adrian Haith, Sethu Vijayakumar

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Many computational models of vestibulo-ocular reflex (VOR) adaptation have been proposed, however none of these models have explicitly highlighted the distinction between adaptation to dynamics transformations, in which the intrinsic properties of the oculomotor plant change, and kinematic transformations, in which the extrinsic relationship between head velocity and desired eye velocity changes (most VOR adaptation experiments use kinematic transformations to manipulate the desired response). We show that whether a transformation is kinematic or dynamic in nature has a strong impact upon the speed and stability of learning for different control architectures. Specifically, models based on a purely feedforward control architecture, as is commonly used in feedback-error learning (FEL), are guaranteed to be stable under kinematic transformations, but are susceptible to slow convergence and instability under dynamics transformations. On the other hand, models based on a recurrent cerebellar architecture [7] perform well under dynamics but not kinematics transformations. We apply this insight to derive a new model of the VOR/OKR system which is stable against transformations of both the plant dynamics and the task kinematics.

Original languageEnglish (US)
Title of host publication2007 IEEE 6th International Conference on Development and Learning, ICDL
Pages37-42
Number of pages6
DOIs
StatePublished - 2007
Externally publishedYes
Event2007 IEEE 6th International Conference on Development and Learning, ICDL - London, United Kingdom
Duration: Jul 11 2007Jul 13 2007

Other

Other2007 IEEE 6th International Conference on Development and Learning, ICDL
CountryUnited Kingdom
CityLondon
Period7/11/077/13/07

Fingerprint

Kinematics
Feedforward control
Feedback
Experiments

ASJC Scopus subject areas

  • Computational Theory and Mathematics
  • Computer Science Applications

Cite this

Haith, A., & Vijayakumar, S. (2007). Robustness of VOR and OKR adaptation under kinematics and dynamics transformations. In 2007 IEEE 6th International Conference on Development and Learning, ICDL (pp. 37-42). [4354055] https://doi.org/10.1109/DEVLRN.2007.4354055

Robustness of VOR and OKR adaptation under kinematics and dynamics transformations. / Haith, Adrian; Vijayakumar, Sethu.

2007 IEEE 6th International Conference on Development and Learning, ICDL. 2007. p. 37-42 4354055.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Haith, A & Vijayakumar, S 2007, Robustness of VOR and OKR adaptation under kinematics and dynamics transformations. in 2007 IEEE 6th International Conference on Development and Learning, ICDL., 4354055, pp. 37-42, 2007 IEEE 6th International Conference on Development and Learning, ICDL, London, United Kingdom, 7/11/07. https://doi.org/10.1109/DEVLRN.2007.4354055
Haith A, Vijayakumar S. Robustness of VOR and OKR adaptation under kinematics and dynamics transformations. In 2007 IEEE 6th International Conference on Development and Learning, ICDL. 2007. p. 37-42. 4354055 https://doi.org/10.1109/DEVLRN.2007.4354055
Haith, Adrian ; Vijayakumar, Sethu. / Robustness of VOR and OKR adaptation under kinematics and dynamics transformations. 2007 IEEE 6th International Conference on Development and Learning, ICDL. 2007. pp. 37-42
@inproceedings{c9f78bb3c32c4751ab05780ff602805c,
title = "Robustness of VOR and OKR adaptation under kinematics and dynamics transformations",
abstract = "Many computational models of vestibulo-ocular reflex (VOR) adaptation have been proposed, however none of these models have explicitly highlighted the distinction between adaptation to dynamics transformations, in which the intrinsic properties of the oculomotor plant change, and kinematic transformations, in which the extrinsic relationship between head velocity and desired eye velocity changes (most VOR adaptation experiments use kinematic transformations to manipulate the desired response). We show that whether a transformation is kinematic or dynamic in nature has a strong impact upon the speed and stability of learning for different control architectures. Specifically, models based on a purely feedforward control architecture, as is commonly used in feedback-error learning (FEL), are guaranteed to be stable under kinematic transformations, but are susceptible to slow convergence and instability under dynamics transformations. On the other hand, models based on a recurrent cerebellar architecture [7] perform well under dynamics but not kinematics transformations. We apply this insight to derive a new model of the VOR/OKR system which is stable against transformations of both the plant dynamics and the task kinematics.",
author = "Adrian Haith and Sethu Vijayakumar",
year = "2007",
doi = "10.1109/DEVLRN.2007.4354055",
language = "English (US)",
isbn = "1424411165",
pages = "37--42",
booktitle = "2007 IEEE 6th International Conference on Development and Learning, ICDL",

}

TY - GEN

T1 - Robustness of VOR and OKR adaptation under kinematics and dynamics transformations

AU - Haith, Adrian

AU - Vijayakumar, Sethu

PY - 2007

Y1 - 2007

N2 - Many computational models of vestibulo-ocular reflex (VOR) adaptation have been proposed, however none of these models have explicitly highlighted the distinction between adaptation to dynamics transformations, in which the intrinsic properties of the oculomotor plant change, and kinematic transformations, in which the extrinsic relationship between head velocity and desired eye velocity changes (most VOR adaptation experiments use kinematic transformations to manipulate the desired response). We show that whether a transformation is kinematic or dynamic in nature has a strong impact upon the speed and stability of learning for different control architectures. Specifically, models based on a purely feedforward control architecture, as is commonly used in feedback-error learning (FEL), are guaranteed to be stable under kinematic transformations, but are susceptible to slow convergence and instability under dynamics transformations. On the other hand, models based on a recurrent cerebellar architecture [7] perform well under dynamics but not kinematics transformations. We apply this insight to derive a new model of the VOR/OKR system which is stable against transformations of both the plant dynamics and the task kinematics.

AB - Many computational models of vestibulo-ocular reflex (VOR) adaptation have been proposed, however none of these models have explicitly highlighted the distinction between adaptation to dynamics transformations, in which the intrinsic properties of the oculomotor plant change, and kinematic transformations, in which the extrinsic relationship between head velocity and desired eye velocity changes (most VOR adaptation experiments use kinematic transformations to manipulate the desired response). We show that whether a transformation is kinematic or dynamic in nature has a strong impact upon the speed and stability of learning for different control architectures. Specifically, models based on a purely feedforward control architecture, as is commonly used in feedback-error learning (FEL), are guaranteed to be stable under kinematic transformations, but are susceptible to slow convergence and instability under dynamics transformations. On the other hand, models based on a recurrent cerebellar architecture [7] perform well under dynamics but not kinematics transformations. We apply this insight to derive a new model of the VOR/OKR system which is stable against transformations of both the plant dynamics and the task kinematics.

UR - http://www.scopus.com/inward/record.url?scp=50849087349&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=50849087349&partnerID=8YFLogxK

U2 - 10.1109/DEVLRN.2007.4354055

DO - 10.1109/DEVLRN.2007.4354055

M3 - Conference contribution

AN - SCOPUS:50849087349

SN - 1424411165

SN - 9781424411160

SP - 37

EP - 42

BT - 2007 IEEE 6th International Conference on Development and Learning, ICDL

ER -