Unified Training of Universal Time Series Forecasting Transformers
This work addresses the limitation of one-model-per-dataset frameworks in time series forecasting, offering a universal model that could benefit practitioners across domains, though it is incremental in building on existing Transformer architectures.
The paper tackles the problem of training a single universal model for diverse time series forecasting tasks by addressing challenges like cross-frequency learning and varying distributions, resulting in Moirai, which achieves competitive or superior zero-shot performance compared to full-shot models on a dataset with over 27B observations.
Deep learning for time series forecasting has traditionally operated within a one-model-per-dataset framework, limiting its potential to leverage the game-changing impact of large pre-trained models. The concept of universal forecasting, emerging from pre-training on a vast collection of time series datasets, envisions a single Large Time Series Model capable of addressing diverse downstream forecasting tasks. However, constructing such a model poses unique challenges specific to time series data: i) cross-frequency learning, ii) accommodating an arbitrary number of variates for multivariate time series, and iii) addressing the varying distributional properties inherent in large-scale data. To address these challenges, we present novel enhancements to the conventional time series Transformer architecture, resulting in our proposed Masked Encoder-based Universal Time Series Forecasting Transformer (Moirai). Trained on our newly introduced Large-scale Open Time Series Archive (LOTSA) featuring over 27B observations across nine domains, Moirai achieves competitive or superior performance as a zero-shot forecaster when compared to full-shot models. Code, data, and model weights can be found at https://github.com/SalesforceAIResearch/uni2ts.