

xLSTM-based time series model TiRex - why is it leading the leaderboards?
Jun 4, 2025
Sepp Hochreiter, a pioneer of LSTM and head of the Institute for Machine Learning at Johannes Kepler University, discusses TiRex, a revolutionary time series model surpassing industry giants like Amazon and Google. He explains the evolution of LSTM to XLSDM, emphasizing its strengths in forecasting. TiRex is designed for user-friendly industrial applications with impressive speed and accuracy. Hochreiter also compares TiRex with competitors and shares insights on job opportunities for those eager to join the innovative team in Linz.
AI Snips
Chapters
Transcript
Episode notes
XLSDM Enhances LSTM for Time Series
- XLSDM combines strengths of LSTM and Transformer models for time-series data.
- It enables parallelization and fast processing while maintaining LSTM's state tracking ability.
Time Series Impact Across Industries
- Time series data pervades domains from weather and stock markets to healthcare and energy.
- Sepp shares examples like flood prediction using hydrology models with hidden states.
LSTM Captures Hidden System States
- LSTM excels at capturing hidden states in complex systems, outperforming traditional statistical models.
- This ability makes it especially suited for predicting time series with underlying states, like hydrology or pipes.