đđ Edge#121: Transformers and Time Series   Â
Learn about TFT, GluonTS and what is the main challenge for applying transformer models to time series forecasting
In this issue:
we discuss transformers and time series;
we explore Google Researchâs paper about temporal fusion transformers;
we overview GluonTS, a time series forecasting framework that includes transformer architectures.
đĄÂ ML Concept of the Day: Transformers and Time Series   Â
The emergence of transformer architectures opened new opportunities to expand the ideas around time series analysis. Specifically, transformers are applicable in multivariate time series forecasting scenarios in which various orthogonal time series need to be combined to produce predictions, which sometimes expand through multiple time horizons. Traditional CNN and RNN models have limitations in those scenarios, given that their complexity increases rapidly with the input size.  Â
đ ML Research You Should Know: Temporal Fusion Transformers Bring Transformer Architectures to Time Series Forecasting
Almost since the release of the initial attention-based transformer models, the researchers attempted to adapt them to the universe of time series. After all, if transformer architectures result in a breakthrough in the time series space, it can unleash an innovation race in areas like quant models in financial markets. The results of such adaptation were mixed. Google Researchâs Temporal Fusion Transformer (TFT) stands out as one of the most solid models implemented in several time series forecasting stacks ->read how Googleâs TFT expands traditional encoder-decoder transformer models for multi-horizon time series forecasting scenarios
đ Letâs connect
Follow us on Twitter. We share useful information, events, educational materials, and informative threads that help you strengthen your knowledge about ML and AI.
đ€ ML Technology to Follow: GluonTS is a Time Series Forecasting Framework that Includes Transformer ArchitecturesÂ
Why should I know about this: GluonTS enables simple time-series forecasting models based on the Apache MxNet framework and is actively used in many of Amazonâs mission-critical applications ->what is it and how you can use it