PatchTST: A Breakthrough in Time Series Forecasting

Transformer-based models have been successfully applied in many fields like natural language processing (think BERT or GPT models) and computer vision to name a few. However, when it comes to time series, state-of-the-art results have mostly been achieved by MLP models (multilayer perceptron) such as <a href="https://medium.com/towards-data-science/the-easiest-way-to-forecast-time-series-using-n-beats-d778fcc2ba60" rel="noopener">N-BEATS</a> and <a href="https://towardsdatascience.com/all-about-n-hits-the-latest-breakthrough-in-time-series-forecasting-a8ddcb27b0d5" rel="noopener" target="_blank">N-HiTS</a>. A recent paper even shows that simple linear models outperform complex transformer-based forecasting models on many benchmark datasets (see <a href="https://arxiv.org/pdf/2205.13504.pdf" rel="noopener ugc nofollow" target="_blank">Zheng et al., 2022</a>). Still, a new transformer-based model has been proposed that achieves state-of-the-art results for long-term forecasting tasks: PatchTST. <a href="https://medium.com/towards-data-science/patchtst-a-breakthrough-in-time-series-forecasting-e02d48869ccc">Read More</a>