Summary of Cross-domain Pre-training with Language Models For Transferable Time Series Representations, by Mingyue Cheng et al.
Cross-Domain Pre-training with Language Models for Transferable Time Series Representations
by Mingyue Cheng, Xiaoyu Tao, Qi Liu, Hao Zhang, Yiheng Chen, Defu Lian
First submitted to arxiv on: 19 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes CrossTimeNet, a novel framework for learning transferable knowledge from various time-series domains to benefit downstream tasks. The main challenge lies in the differences between domains, such as varying number of channels and temporal resolution scales. To address this, CrossTimeNet incorporates a newly designed time series tokenization module, which optimizes reconstruction to convert raw time series into discrete tokens. Additionally, predicting corrupted tokens can extract informative patterns across domains during self-supervised learning (SSL) pre-training, often overlooked in previous works. The paper also treats the pre-training language model (PLM) as the initialization of the encoder network, exploring knowledge transfer from PLM to time series area. Experimental results confirm CrossTimeNet’s superior performance in real-world scenarios across various time series classification domains. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper creates a way to learn from many different types of time-series data, which is very useful for improving tasks that use this type of data. The main challenge was the differences between these types of data, such as how they’re organized and what kind of information they contain. To solve this, the researchers developed a new way to convert raw time series into tokens that can be used by computers. They also found that predicting errors in these tokens can help identify patterns across different datasets during training. The paper also shows that knowledge learned from language models can be transferred to time-series data, which is a new and exciting area of research. |
Keywords
* Artificial intelligence * Classification * Encoder * Language model * Self supervised * Time series * Tokenization
