Language Pre-training Guided Masking Representation Learning for Time Series Classification

Authors: Liaoyuan Tang, Zheng Wang, Jie Wang, Guanxiong He, Zhezheng Hao, Rong Wang, Feiping Nie

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the performance of proposed representation learning via classification task conducted on 106 time series datasets, which demonstrates the effectiveness of proposed method. In this section, to demonstrate the effectiveness of LPMRL, we first perform a comparison with the baseline algorithm on 106 datasets from the UCR time series dataset. Second, we perform ablation experiments on the model to evaluate the contribution of each module within the model. Third, visualization experiments are conducted to validate the classification performance of LPMRL, focusing on its representational dimension.
Researcher Affiliation Academia Liaoyuan Tang, Zheng Wang*, Jie Wang, Guanxiong He, Zhezheng Hao, Rong Wang, Feiping Nie School of Artificial Intelligence, Optics and Electronics (i OPEN), Northwestern Polytechnical University 127 West Youyi Road, Beilin District Xi an Shaanxi, 710072, P.R.China EMAIL, EMAIL, EMAIL, EMAIL, EMAIL, EMAIL, EMAIL
Pseudocode No The paper describes the proposed method, Language Guided Masking Autoencoder and Dual-information Contrastive Learning, using prose and mathematical equations. It does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statements about releasing source code, nor does it include links to a code repository.
Open Datasets Yes Our approach was rigorously evaluated using the well-established UCR time series archive (Dau et al. 2019), a benchmark dataset in realworld scenarios.
Dataset Splits No The paper states: "Our approach was rigorously evaluated using the well-established UCR time series archive (Dau et al. 2019)... Out of the 128 available UCR datasets, we carefully chose 106 datasets... We randomly selected a certain ratio of the 106 datasets in which multiple algorithms were compared." While it mentions evaluating on a percentage of datasets and adhering to TS2Vec's setup for an SVM classifier, it does not provide specific train/test/validation splits (e.g., percentages or sample counts) for individual datasets needed to reproduce the data partitioning.
Hardware Specification Yes These experiments were meticulously conducted in a controlled environment using Py Torch 1.10, running on two high-performance NVIDIA Ge Force RTX A6000 GPUs, ensuring that our results were both reliable and reproducible.
Software Dependencies Yes These experiments were meticulously conducted in a controlled environment using Py Torch 1.10, running on two high-performance NVIDIA Ge Force RTX A6000 GPUs, ensuring that our results were both reliable and reproducible.
Experiment Setup Yes To maintain consistency across methods, we configured the representation dimension for all classification techniques, except for DTW, to 320. Additionally, the maximum epoch was set at 500, the learning rate was configured to 1e-4, and the batch size was set to 64.