Developmental learning of memory-based perceptual models

This paper reviews an online learning algorithm for incremental learning of memory-based utterance models. The simple core algorithm is augmented with a strategy for selecting the compression set-a subset of the input data that possesses some optimality characteristics. These sets are again found incrementally. Several strategies are formulated and empirically compared on three standard data sets. The algorithm is used as a perceptual learning component of an adaptive autonomous agent.