L² 距離推定モデルにおけるパラメータ探索のための蛍アルゴリズムの適用

抄録

When analyzing uncertain data expressed by a set of data points, calculating the density difference between any two sets of data points is an important task. Least squares density difference (LSDD) estimator which uses a non-parametric model approximates the true density difference accurately. In the previous method, it searches for the parameters of the model by using the grid search. However, there is no guarantee that the obtained parameters by the grid search are optimal, we apply a firefly algorithm which is suitable to the multimodal optimization problem for the parameter search. Numerical experiments are conducted to show that the LSDD estimator by using the firefly algorithm approximate the true density differences and L²-distances more accurately than the grid search.