Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]

Frequency-Masked Embedding Inference: A Non-Contrastive Approach for Time Series Representation Learning

Authors: En Fu, Yanyan Hu

AAAI 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on 8 widely used time series datasets for classification and regression tasks, using linear evaluation and end-to-end fine-tuning, show that FEI significantly outperforms existing contrastive-based methods in terms of generalization.
Researcher Affiliation Academia En Fu1, Yanyan Hu1,2,* 1School of Intelligence Science and Technology, University of Science and Technology Beijing 2Institute of Artificial Intelligence, University of Science and Technology Beijing EMAIL, EMAIL
Pseudocode No The paper describes the methods and processes using mathematical equations and descriptive text but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code/Appendix https://github.com/USTBInnovationPark/Frequencymasked-Embedding-Inference
Open Datasets Yes The dataset used in the experimental section is shown in Table 1. For the pre-training phase, we use the commonly utilized SLeep EEG dataset, which provides ample samples and is widely used for pre-training in various transfer learning methods (Zhang et al. 2022; Dong et al. 2023). For the downstream tasks, 8 publicly available datasets are utilized: 1) Gesture (Liu et al. 2009), 2) FD-B (Lessmeier et al. 2016), 3) EMG (Goldberger et al. 2000), 4) EPI(Andrzejak et al. 2001), 5) HAR (Anguita et al. 2013), 6) 128 UCR (Dau et al. 2018), 7) C-MAPSS (Saxena et al. 2008), and 8) Bearing (Wang et al. 2018). The UCR Time Series Classification Archive. https://www.cs.ucr.edu/ eamonn/time series data 2018/. Accessed: 2024-08-01.
Dataset Splits Yes The model with the lowest validation loss is used as the final test model by early stopping for all datasets except for the 128 UCR dataset, which does not have a validation set division, so all models are tested directly after training ends.
Hardware Specification No The paper discusses model training and evaluation but does not specify any particular hardware (e.g., GPU models, CPU types) used for the experiments.
Software Dependencies No The paper mentions using '1D Res Net' as the encoder network but does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow, CUDA versions).
Experiment Setup Yes In the linear evaluation process, we freeze the encoder, and optimize only a linear classifier, with a maximum training iteration of 300 and an initial learning rate of 1e-4. In the end-to-end fine-tuning process, both the encoder and the classifier are optimized, with a maximum iteration of 100 and a smaller learning rate of 1e-5 to prevent overfitting. The basic setup of proposed FEI for the pre-training process follows Sim MTM and TF-C. Additional configurations can be found in the Appendix.