Analyzing movement trajectories using a Markov bi-clustering method.

Area of Science:

• Computational Neuroscience
• Cognitive Science
• Robotics

Background:

• Scribbling is a fundamental human motor activity.
• Previous models often treat scribbles as monolithic entities.
• Understanding the compositional nature of scribbling can unlock insights into motor control and learning.

Purpose of the Study:

• To model scribbling motion as a compositional system.
• To develop a method for hierarchically decomposing complex scribbles into elementary units.
• To analyze the structure of human scribbling using information theory and machine learning.

Main Methods:

• Treating scribbling as a compositional system of concatenating elementary strokes.
• Segmenting continuous scribbles into discrete units and calculating Markovian transition matrices between trajectory clusters.
• Applying a novel Markov-state bi-clustering algorithm based on the Information-Bottleneck principle to group states and minimize mutual information loss.
• Hierarchically decomposing scribblings into increasingly finer elements.

Main Results:

• Demonstrated that scribbling can be effectively represented as a compositional system.
• Developed and applied a novel algorithm for hierarchical decomposition of scribbles.
• Successfully illustrated the algorithm's utility by applying it to human scribbling data.
• Identified underlying structural patterns in complex scribbling behaviors.

Conclusions:

• Scribbling motion can be understood as a generative system built from a finite set of basic strokes.
• The developed information-theoretic approach provides a powerful tool for analyzing complex sequential data.
• This hierarchical decomposition method offers new avenues for studying motor control, learning, and potentially for applications in human-computer interaction and artificial intelligence.