%0 Conference Paper %1 Haruno:2003 %A Haruno, Masahiko %A Wolpert, Daniel M. %A Kawato, Mitsuo %B Excepta Medica International Coungress Series %C Amsterdam, The Netherlands %D 2003 %E Ono, T. %E Matsumoto, G. %E Llinas, R.R. %E Berthoz, A. %E Norgren, R. %E Nishijo, H. %E Tamura, R. %I Elsevier Science B.V. %K Cerebellum; Cerebro? Hierarchy; Internal MOSAIC; Predictors and cerebellar communication controllers; loop model; %P 575-590 %R 10.1016/S0531-5131(03)00190-0 %T Hierarchical MOSAIC for movement generation %V 1250 %X Hierarchy plays a key role in human motor control and learning. We can generate a variety of structured motor sequences such as writing or speech and learn to combine elemental actions in novel orders. We previously proposed the Modular Selection and Identification for Control (MOSAIC) model to explain the remarkable ability animals show in motor learning, adaptation and behavioral switching. In this paper, we extend this to a hierarchical MOSAIC (HMOSAIC). Each layer of HMOSAIC consists of a MOSAIC, which is a set of paired control and predictive models. The higher-level MOSAIC receives two inputs: an abstract (symbolic) desired trajectory and posterior probabilities of its subordinate level, which represent which modules are playing a crucial role in the lower level under the current behavioral situation. The higher control model generates, as a motor command, prior probabilities for the lower-level modules, and therefore prioritizes which lower-level modules should be selected. In contrast, the higher predictive model learns to estimate the posterior probability at the next time step. The outputs from controllers as well as the learning of both predictors and controllers are weighted by the precision of the prediction. We first show that this bidirectional architecture provides a general framework capable of hierarchical motor learning that is chunking of movement patterns. Then, we discuss the similarities between the HMOSAIC architecture and the closed cerebro?cerebellar loop circuits recently found by Middleton and Strick (Trends in Neuroscience 21 (1998) 367). In our view, modules in one layer are involved with similar functions and assumed to be implemented by one of the cerebro?cerebellar loop circuits. These layers are then connected to each other by the bidirectional information flows within the cerebral cortex.

@inproceedings{Haruno:2003, abstract = {Hierarchy plays a key role in human motor control and learning. We can generate a variety of structured motor sequences such as writing or speech and learn to combine elemental actions in novel orders. We previously proposed the Modular Selection and Identification for Control (MOSAIC) model to explain the remarkable ability animals show in motor learning, adaptation and behavioral switching. In this paper, we extend this to a hierarchical MOSAIC (HMOSAIC). Each layer of HMOSAIC consists of a MOSAIC, which is a set of paired control and predictive models. The higher-level MOSAIC receives two inputs: an abstract (symbolic) desired trajectory and posterior probabilities of its subordinate level, which represent which modules are playing a crucial role in the lower level under the current behavioral situation. The higher control model generates, as a motor command, prior probabilities for the lower-level modules, and therefore prioritizes which lower-level modules should be selected. In contrast, the higher predictive model learns to estimate the posterior probability at the next time step. The outputs from controllers as well as the learning of both predictors and controllers are weighted by the precision of the prediction. We first show that this bidirectional architecture provides a general framework capable of hierarchical motor learning that is chunking of movement patterns. Then, we discuss the similarities between the HMOSAIC architecture and the closed cerebro?cerebellar loop circuits recently found by Middleton and Strick (Trends in Neuroscience 21 (1998) 367). In our view, modules in one layer are involved with similar functions and assumed to be implemented by one of the cerebro?cerebellar loop circuits. These layers are then connected to each other by the bidirectional information flows within the cerebral cortex.}, added-at = {2009-06-26T15:25:19.000+0200}, address = {Amsterdam, The Netherlands}, author = {Haruno, Masahiko and Wolpert, Daniel M. and Kawato, Mitsuo}, biburl = {https://www.bibsonomy.org/bibtex/2b4479089d98a6241cff44e6f5c2f6bd7/butz}, booktitle = {Excepta Medica International Coungress Series}, comment = {Im Mosaic werden die IMs und FMs genutzt um das Verhalten des Plant zu kontrollieren bzW. vorherzusagen. In der nächsten Hierarchiestufe wird nicht mehr das Plant kontrolliert sondern die Responsibilities der darunter liegenden Ebene. Dabei liefern die unteren Module die Tatsächliche Reponsibility während die höheren Ebenen versuche vorherzusagen welche Module wohl responsible sein werden und diese dann auch voraktivieren. Dieses Konzept kann zur Wahrnehmung einer abstrakteren Situation (Trajektorie, Aktionen) etc. verwendet werden und diese auch kontrollieren. Anstelle der Modelle könne in höheren Ebenen auch Pattern Generator oder Reinforcement learning controller eingebaut werden. Ein Zusammenhang zu motorischen Zentren im Gehirn wird hergestellt. Getestet wird es mit einer Trackingaufgabe, wobei sich die Umwelt schrittweise verändert. Die Hierarchie kann auch komplette leichte Veränderungen der Umwelt, die nicht durch ein LL-Modul repräsentiert sind, abfangen.}, description = {diverse cognitive systems bib}, doi = {10.1016/S0531-5131(03)00190-0}, editor = {Ono, T. and Matsumoto, G. and Llinas, R.R. and Berthoz, A. and Norgren, R. and Nishijo, H. and Tamura, R.}, interhash = {6e9743054e23c72705d170dce5dac9bc}, intrahash = {b4479089d98a6241cff44e6f5c2f6bd7}, keywords = {Cerebellum; Cerebro? Hierarchy; Internal MOSAIC; Predictors and cerebellar communication controllers; loop model;}, owner = {martin}, pages = {575-590}, publisher = {Elsevier Science B.V.}, timestamp = {2009-06-26T15:25:33.000+0200}, title = {Hierarchical MOSAIC for movement generation}, volume = 1250, year = 2003 }