Are Transformers Effective for Time Series Forecasting?
- Time series are ubiquitous in today’s data-driven world.
- Given historical data, time series forecasting (TSF) is a long-standing task that has a wide range of applications
- Long-term time series forecasting (LTSF)
- Simple one-layer linear models named LTSF-Linear for comparison
- The main working power of Transformers is from its multi-head self-attention mechanism
- Extensive experiments on nine widely-used benchmark datasets that cover various real-life applications: traffic, energy, economics, weather, and disease predictions.
- LTSF-Linear outperforms existing complex Transformerbased models in all cases, and often by a large margin (20% - 50%).
- The difference between the original sequence and the trend component is regarded as the seasonal component
- Transformer-based methods: FEDformer [31], Autoformer [28], Informer [30], Pyraformer [18], and LogTrans [16].
- FEDformer in most cases by 20% ∼ 50% improvements on the multivariate forecasting
- FEDformer employs classical time series analysis techniques such as frequency processing, which brings in time series inductive bias and benefits the ability of temporal feature extraction
- Recurrent neural networks (RNNs) based methods (e.g., [21]) summarize the past information compactly in internal memory states and recursively update themselves for forecasting.
- Convolutional neural networks (CNNs) based methods (e.g., [3]), wherein convolutional filters are used to capture local temporal features.
Keep Exploring!!!
No comments:
Post a Comment