Processing Markov Logic Networks with GPUs
Departamento de Computacion, CINVESTAV-IPN
25th International Conference On Inductive Logic Programming, 2015
@article{martinez2015processing,
title={Processing Markov Logic Networks with GPUs},
author={Mart{i}nez-Angeles, Carlos Alberto and Dutra, In{^e}s and Costa, V{i}tor Santos and Buenabad-Ch{‘a}vez, Jorge},
year={2015}
}
Graphics Processing Units (GPUs) are being widely used to improve performance of machine learning and logic programming systems. Next, we propose using this technique to improve the performance of Markov logic programs. In this paper we focus on the first step of the inference phase, the grounding of first-order logical formulas composing a Markov network. Our system, Tu2GPU, is based on the state-of-theart high-performance Tuffy system. We compare Tu2GPU’s performance to that of the Alchemy, Tuffy and RockIt systems on three widely used applications. Results show that Tu2GPU is up to 15x times faster than the other systems.
August 14, 2015 by hgpu
