Time series foundation models

11:45-13:15 · Harsha Halgamuwe Hewage

Note

Slides for this session are still being prepared and will be added closer to the workshop.

This new session introduces time series foundation models: large, pre-trained models (such as Chronos, TimeGPT and Moirai) that can produce forecasts directly from historical data with little or no task-specific training. It will cover how these models work at a high level, how they compare to the classical methods covered earlier in the workshop, and when they are (and aren’t) a good choice for forecasting problems in low- and middle-income country contexts.