Premium
Key Time Steps Selection for Large‐Scale Time‐Varying Volume Datasets Using an Information‐Theoretic Storyboard
Author(s) -
Zhou Bo,
Chiang YiJen
Publication year - 2018
Publication title -
computer graphics forum
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.578
H-Index - 120
eISSN - 1467-8659
pISSN - 0167-7055
DOI - 10.1111/cgf.13399
Subject(s) - computer science , preprocessor , key (lock) , interpolation (computer graphics) , time complexity , data mining , selection (genetic algorithm) , storyboard , algorithm , machine learning , artificial intelligence , image (mathematics) , multimedia , computer security
Abstract Key time steps selection is essential for effective and efficient scientific visualization of large‐scale time‐varying datasets. We present a novel approach that can decide the number of most representative time steps while selecting them to minimize the difference in the amount of information from the original data. We use linear interpolation to reconstruct the data of intermediate time steps between selected time steps. We propose an evaluation of selected time steps by computing the difference in the amount of information (called information difference ) using variation of information (VI) from information theory, which compares the interpolated time steps against the original data. In the one‐time preprocessing phase, a dynamic programming is applied to extract the subset of time steps that minimize the information difference. In the run‐time phase, a novel chart is used to present the dynamic programming results, which serves as a storyboard of the data to guide the user to select the best time steps very efficiently. We extend our preprocessing approach to a novel out‐of‐core approximate algorithm to achieve optimal I/O cost, which also greatly reduces the in‐core computing time and exhibits a nice trade‐off between computing speed and accuracy. As shown in the experiments, our approximate method outperforms the previous globally optimal DTW approach [TLS12] on out‐of‐core data by significantly improving the running time while keeping similar qualities, and is our major contribution.