Methods for Quantifying the Informational Structure of Sensory and Motor Data

Statistical regularities in the sensory data relayed to the brain are critical for enabling appropriate developmental processes, perceptual categorization, adaptation, and learning. For robots such regularities are important because they avoid the allocation of potentially limited processing resources if not strictly necessary. There are at least three alternative mechanisms by which informational structure is produced in the sensory channels of natural organisms and robotic artifacts: (1) clever pre-processing strategies (computational mechanism), (2) an adequate bodily setup (morphological mechanism), and (3) appropriate behavioral interaction.

The aim of this project is to shed light on the third (behavioral) mechanism. The core idea is to quantitatively characterize the informational structure of sensory and motor data by means of a set of univariate and multivariate statistical and information-theoretical measures. More specifically, we propose a MatLab toolbox comprising functions for density estimation, entropy, mutual information, complexity, linear and nonlinear dimensionality reduction, and so on.

All the functions have been implemented in MatLab, but should be easily portable to Octave. The OS platforms supported are all variants of Windows and Linux. A detailed illustration of the toolbox can be also found in a scientific paper which is under submission (available upon request). The potential importance of this project is to further our understanding of processes of sensorimotor coordination in organisms and for robot design. We also expect to apply the proposed and similar methods to the design of robotic artifacts.