A Fast Partial Memory Approach to Incremental Learning through an Advanced Data Storage Framework.
Marenglen Biba, Stefano Ferilli, Floriana Esposito, Nicola Di Mauro, Teresa M. A. Basile · 2007
Abstract. Inducing concept descriptions from examples has been thoroughly tackled by symbolic machine learning methods. However, on-line learning methods that acquire concepts from examples distributed over time, require great computational effort. This is not only due to the intrinsic complexity of the concept learning task, but also to the full memory approach that most learning systems adopt. Indeed, during learning, most of these systems consider all their past examples leading to expensive procedures for consistency verification. In this paper, we present an implementation of a partial memory approach through an advanced data storage framework and show through experiments that great savings in learning times can be achieved. We also propose and experiment different ways to select the past examples paving the way for further research in on-line partial memory learning agents. 1