Supervised Contrastive Few-Shot Learning for High-Frequency Time Series
Authors: Xi Chen, Cheng Ge, Ming Wang, Jin Wang
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on four mainstream public datasets as well as a series of sensitivity and ablation studies demonstrate that the learned representations are effective and robust compared with the direct supervised learning and self-supervised learning, notably under the minimal few-shot situation. |
| Researcher Affiliation | Industry | Xi Chen, Cheng Ge, Ming Wang, Jin Wang Alibaba Group {chuyu.cx, eric.gc}@alibaba-inc.com, duchen.wm@taobao.com, jet.wangj@alipay.com |
| Pseudocode | No | The paper provides architectural diagrams and mathematical formulations but does not include pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statements or links to open-source code for the described methodology. |
| Open Datasets | Yes | MFPT This dataset is provided by Society of Machinery Failure Prevention Technology (MFPT 2018). PU The bearing dataset is from Paderborn University (PU 2016). CWRU This is one of the most widely used datasets provided by Case Western Reserve University Bearing Data Center (CWRU 2015). SEU This dataset has both gearbox and bearing vibrations provided by Southeast University (SEU 2018). |
| Dataset Splits | No | Due to the very limited number of training samples, the training set is taken as a whole without further splitting into train and validation. |
| Hardware Specification | Yes | Considering training uncertainties, all experiments are repeated ten times on a Mac Book Pro with 2.4GHz processor and 16GB RAM. |
| Software Dependencies | No | The paper does not provide specific version numbers for software dependencies used in the experiments. |
| Experiment Setup | Yes | The parameter setting for frequency input is recommended as follows: the channel sizes and the kernel numbers of four 1-d convolutional extractors are 16, 32, 64, 128 and 15, 9, 7, 5 respectively. The sizes of Max Pool and Adaptive Pool are 2 and 64. The output dimension of the fully-connected layer is set to 64. Let τ is a scalar temperature parameter for tuning how concentrated the features are in the representation space. For a weight factor α [0, 1], the overall loss function is expressed as follows: Experiments are conducted under two situations where Nsample = 4 and Nsample = 8. For each batch, the number of samples with the same label is denoted as Nview. To maintain the same number of batches in two situations, Nview is set to 2 and 4, respectively. |