TimeFound: A Foundation Model for Time Series Forecasting
By: Congxi Xiao , Jingbo Zhou , Yixiong Xiao and more
Potential Business Impact:
Predicts future events without prior training.
We present TimeFound, an encoder-decoder transformer-based time series foundation model for out-of-the-box zero-shot forecasting. To handle time series data from various domains, TimeFound employs a multi-resolution patching strategy to capture complex temporal patterns at multiple scales. We pre-train our model with two sizes (200M and 710M parameters) on a large time-series corpus comprising both real-world and synthetic datasets. Over a collection of unseen datasets across diverse domains and forecasting horizons, our empirical evaluations suggest that TimeFound can achieve superior or competitive zero-shot forecasting performance, compared to state-of-the-art time series foundation models.
Similar Papers
Foundation Models for Time Series: A Survey
Machine Learning (CS)
Helps computers understand patterns in data over time.
Benchmarking Time Series Forecasting Models: From Statistical Techniques to Foundation Models in Real-World Applications
Machine Learning (CS)
Predicts restaurant sales using smart computer programs.
Generalisation Bounds of Zero-Shot Economic Forecasting using Time Series Foundation Models
Machine Learning (CS)
Predicts economy without needing old data.