Scott Albert
A tool to fit state-space models of adaptation to sensorimotor data
Expectation Maximization (EM) is a statistical algorithm that can estimate model parameters for a data set containing latent variables. The EM algorithm identifies parameters that maximize the likelihood function for a model of learning where both movement and planning of movement are noisy processes. This technique contrasts with a minimization of squared error approach to model fitting (least-mean-squared-error or LMSE) which is the maximum likelihood estimator for a system that only has noise in the execution of movement.
The mathematical derivation of the theory, and its application to data, are described here.
Here we provide code that uses EM to fit a two-state model of learning to sensorimotor data. This tool, written by Scott Albert, uncovers the hidden fast and slow states of learning from observed behavior. For reference we also provide code that can be used for fitting with LMSE.
For each package, download the zip file and refer to the README file.
We provide two packages for the EM algorithm:
1. This toolbox uses EM to fit sensorimotor data without set breaks (EM.v1.1.zip).
2. This toolbox uses EM to fit sensorimotor data with set breaks (EM.v1.2.zip).
We provide two packages for the LMSE algorithm:
1. This toolbox uses LMSE to fit sensorimotor data without set breaks (LMSE.v1.1.zip).
2. This toolbox uses LMSE to fit sensorimotor data with set breaks (LMSE.v1.2.zip).
If you encounter any issues with these toolboxes, please contact Scott at: salbert8@jhu.edu.

Software and tools