Unsupervised Time-Series Representation Learning with Iterative Bilinear Temporal-Spectral Fusion

Authors: Ling Yang, Shenda Hong

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Specifically, we firstly conducts downstream evaluations on three major tasks for time series including classification, forecasting and anomaly detection. Experimental results shows that our BTSF consistently significantly outperforms the state-of-the-art methods.
Researcher Affiliation Academia 1National Institute of Health Data Science, Peking University, Beijing, China 2Institute of Medical Technology, Health Science Center of Peking University, Beijing, China.
Pseudocode No The paper describes the method and its components mathematically and textually but does not include a formally labeled pseudocode block or algorithm.
Open Source Code No The paper does not provide a link or an explicit statement about the availability of its source code. It only mentions implementing existing works using public codes.
Open Datasets Yes We evaluate our learned representation on downstream classification tasks for time series on widely-used time series classification datasets (Anguita et al., 2013; Goldberger et al., 2000; Andrzejak et al., 2001; Moody, 1983).
Dataset Splits Yes In the training stage, we keep the original train/test splits of datasets and use the training set to train all the models.
Hardware Specification Yes In all experiments, we use Pytorch 1.8.1 (Paszke et al., 2017) and train all the models on a Ge Force RTX 2080 Ti GPU with CUDA 10.2.
Software Dependencies Yes In all experiments, we use Pytorch 1.8.1 (Paszke et al., 2017) and train all the models on a Ge Force RTX 2080 Ti GPU with CUDA 10.2.
Experiment Setup Yes We apply an Adam optimizer (Kingma & Ba, 2017) with a learning rate of 3e-4, weight decay of 1e-4 and batch size is set to 256. ... the dropout rate, temperature number τ and the loops number of iterative bilinear fusion. Table 6 illustrates that when the rate is set to 0.1 ... Table 7 demonstrates that when τ is set to 0.05