Descriptive approach to modelling learning

Martinez-Alvarez, F. Smeraldi and T. Rolleke, in Proceedings of the Spanish Conference on Information Retrieval, 2010

Abstract

The importance of Learning Algorithms in Computer Science and other fields has been increasing in the last years. On the other hand, Descriptive Approaches have significantly impacted different domains, being SQL for data access and management one of the main examples. This paper investigates a descriptive approach for learning.

In the field of IR, learning techniques and descriptive approaches have already been applied independently. For the former, one of the most significant examples is the “Learning to Rank” task, while Probabilistic Datalog, which is a representative of descriptive approaches, has been applied for solving different IR-tasks, providing a high level representation of search strategies.

The main contributions of this paper are a knn classifier implemented in PDatalog, showing that the expressiveness of 2nd generation Probabilistic Datalog is sufficient for modelling lazy-learners, its evaluation
for text classification on the Reuters-21578 collection, and a descriptive modelling of polynomial Mercer Kernels.

Full paper (PDF)

Backlinks: Publications