The switching state-space model (SSSM) [1] is a principled Bayesian model that frames motor learning as online state and parameter estimation in a switching state-space model. State inference is implemented using the generalised pseudo-Bayesian estimator of order 1 (GPB1), which is an assumed density filtering method that approximates the exact posterior of each state (a mixture of Gaussians with ct components, where c is the number of contexts and t is the number of trials) with a single Gaussian. To learn the cue emission probabilities, an online formulation of expectation maximisation is used that can be interpreted as a stochastic approximation recursion on the expected complete-data sufficient statistics.
- Heald, J.B., Ingram, J.N., Flanagan, J.R. & Wolpert, D.M. Multiple motor memories are learned to control different points on a tool. Nature Human Behaviour 2(4), 300-311 (2018). [SharedIt link]
- Download the SSSM.m file.
Feel free to e-mail me at jamesbheald@gmail.com if you have any questions.
The SSSM model is released under the terms of the GNU General Public License v3.0.