Boundary-based lower-bound functions for dynamic time warping and their indexing

Piecewise aggregate representations and lower-bound distance functions for multivariate time series

“…Its performance also depends on a warping window, which means that the corner-like boundaries are chosen from the warping window. The experimental results [28] demonstrated that in most cases LB_ECorner is better than LB_Keogh. Lately, we proposed the extensions of LB_Kim and LB_Keogh and discussed the relationships of the existing lower-bound functions [29].…”

“…An improved PAA [35] considering standard variance of each segment was proposed, and the corresponding function was also lower-bound on Euclidean distance. And more importantly, some efficient distance functions [3,13,24,[26][27][28] were lower-bound on DTW. They are very suitable for time series similarity search and indexing by DTW [4,9].…”

“…It also showed that pruning candidates left from the Zhu-Shasha R * -tree with the full LB_Keogh alone were enough to significantly boost the speed and pruning power. LB_ECorner [28] is based on cornerlike boundaries in warping matrix. Its performance also depends on a warping window, which means that the corner-like boundaries are chosen from the warping window.…”

“…Meanwhile, some distance functions that satisfy the lower-bound condition for dynamic time warping are proposed. In this way, we can use the lower-bound functions to index MTS as the traditional ones [3,24,[26][27][28][29] index univariate time series. Especially, the traditional methods only focus on the comparison between two time series of which the length is equal.…”

Set-based Similarity Search for Time Series

“…Thus, by excluding the series that cannot be in k-NN, LCSS is accelerated. Similarly, DTW is quickly evaluated by its lower-bound functions, which cost less, [35] and some indexing techniques [26]. These acceleration strategies mainly depend on the rapid estimation of accurate distance, which is related to the specific data.…”