Adversarial Dynamic Shapelet Networks

Authors: Qianli Ma, Wanqing Zhuang, Sen Li, Desen Huang, Garrison Cottrell5069-5076

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments conducted on extensive time series data sets show that ADSN is state-of-the-art compared to existing shapelet-based methods. The visualization analysis also shows the effectiveness of dynamic shapelet generation and adversarial training. We conduct experiments on the 85 UCR (Chen et al. 2015) and 8 UEA (Hills et al. 2014) time series datasets.
Researcher Affiliation Academia 1School of Computer Science and Engineering, South China University of Technology, Guangzhou, China 2Department of Computer Science and Engineering, University of California, San Diego, CA, USA
Pseudocode Yes The pseudo code of ADSN is shown in section F of the supplementary material.
Open Source Code Yes The supplementary material mentioned in this paper is available on github1. 1https://github.com/qianlima-lab/ADSN.
Open Datasets Yes We conduct experiments on the 85 UCR (Chen et al. 2015) and 8 UEA (Hills et al. 2014) time series datasets. The statistics of these 26 datasets are shown in section A of the supplementary material.
Dataset Splits Yes Each data set was split into training and testing set using the standard split. The hyper-parameters of ADSN are tuned through a grid search approach based on cross validation.
Hardware Specification Yes The experiments are run on the Tensor Flow platform using an Intel Core i7-6850K, 3.60-GHz CPU, 64-GB RAM and a Ge Force GTX 1080-Ti 11G GPU.
Software Dependencies No The paper mentions "Tensor Flow platform" and "The Adam (Kingma and Ba 2014) optimizer" but does not specify their version numbers.
Experiment Setup Yes λdiv and λadv are set to 0.01 and 0.05, respectively. The hyper-parameters of ADSN are tuned through a grid search approach based on cross validation. The number of shapelets is chosen from k {30, 60, 90, 120}. The dropout rate applied to the softmax layer is evaluated over {0, 0.25, 0.5}. We choose the shapelet lengths according to the length of the time series. ... The Adam (Kingma and Ba 2014) optimizer is employed with an initial learning rate of 0.001.