Skip to main content

xLSTM and TiRex for Time Series Forecasting

arXivGitHubHugging FacePyPI

TiRex is a 35M parameter pre-trained time series forecasting model based on xLSTM.

Key Facts:​

  • Zero-Shot Forecasting: TiRex is a pre-trained model that performs time series forecasting without requiring any training on your data. Simply download and use it.

  • Quantile Predictions: TiRex provides both point estimates and quantile estimates.

  • State-of-the-art Performance over Long and Short Horizons: TiRex achieves top scores in various time series forecasting benchmarks, see GiftEval and ChronosZS. These benchmark show that TiRex provides great performance for both long and short-term forecasting.

Brief Background​

xLSTM (Extended Long Short-Term Memory) is a powerful new architecture for sequence modeling, building upon the success of traditional LSTMs. It introduces key innovations that enhance its ability to capture long-range dependencies and complex patterns in data, making it particularly well-suited for tasks like time series forecasting.

TiRex leverages the power of xLSTM to provide a state-of-the-art framework for time series forecasting. By utilizing the advanced sequence modeling capabilities of xLSTM, TiRex can generate accurate forecasts across various time horizons and data characteristics — without being trained on task-specific data. This capability is referred to zero-shot forecasting.

Potential Applications​

TiRex can be applied to a wide range of real-world time series forecasting problems, including (but not limited to):

  • Demand Forecasting: Estimating future product demand for inventory management and supply chain optimization.
  • Energy Forecasting: Predicting electricity consumption or renewable energy generation.
  • Traffic Forecasting: Anticipating traffic flow for urban planning and navigation systems.
  • Environmental Forecasting: Modeling climate patterns or predicting pollution levels.