Concept Formation Using Incremental Gaussian Mixture Models

Abstract

This paper presents a new algorithm for incremental concept formation based on a Bayesian framework. The algorithm, called IGMM (for Incremental Gaussian Mixture Model), uses a probabilistic approach for modeling the environment, and so, it can rely on solid arguments to handle this issue. IGMM creates and continually adjusts a probabilistic model consistent to all sequentially presented data without storing or revisiting previous training data. IGMM is particularly useful for incremental clustering of data streams, as encountered in the domain of moving object trajectories and mobile robotics. It creates an incremental knowledge model of the domain consisting of primitive concepts involving all observed variables. Experiments with simulated data streams of sonar readings of a mobile robot shows that IGMM can efficiently segment trajectories detecting higher order concepts like “wall at right” and “curve at left”.KeywordsConcept FormationIncremental LearningUnsupervised LearningBayesian MethodsEM AlgorithmFinite MixturesClustering